International Journal of Intelligent Information and Database Systems (57 papers in press)

Regular Issues

• Triple attention-enhanced transformer-based federated meta-learning for epileptic seizure detection  
  by Ashwini Patil, Megharani Patil 
  Abstract: In neurological healthcare, accurately identifying seizure occurrences from the electroencephalogram (EEG) signals is essential, implying the epileptic seizure detection task. data privacy, epileptic knowledge scarcity, non-independent and identically distributed (non-IID) characteristics, and inter-patient variability pose research constraints to the traditional centralised learning systems. To develop a reliable and private-preserving model for patient-specific knowledge-aware epileptic seizure detection, this paper suggests a unique model that combines meta-learning with federated learning (FL). The proposed approach applies the ternary feature extraction and hybrid augmentation methods to enhance the comprehensive learning of EEG features over data scarcity. Subsequently, the design of the transformer-based model in federated meta-learning architecture significantly captures the intricate relationships with long-range dependencies in the sequential EEG signals of each federated client while performing task-specific learning within each local EEG data. Thus, the stacked transformer encoder with triple attention in the local model inherently learns the discriminative ictal and non-ictal EEG patterns with the updates of patient-specific learning through the collaborative training across the patients by meta-learning and federated clients by FL, improving the epileptic seizure detection performance.
  Keywords: epileptic seizure detection; ternary feature extraction; hybrid augmentation; stacked transformer encoder with triple attention; federated learning; FL; meta-learning; transformer; and multi-head self-attention.
  DOI: 10.1504/IJIIDS.2025.10072488
   
  
• Enhanced imputation genetic algorithm: a novel approach for data intelligent imputation  
  by V. Amala Deepa, T. Lucia Agnes Beena 
  Abstract: The imputation of missing data in multivariate datasets has been used to enhance the accuracy and reliability of statistical analyses and the machine learning model, especially when the integrity of data can directly impact their decisions; healthcare and finance are basic examples. Methods become biased or inaccurate in these traditional imputation methods as they are not complex enough for multivariate data. It introduces an entirely novel data imputation method called the enhanced imputation genetic algorithm. Such enables dynamic control over genetic operators with the means that crossover and mutation rates may contribute toward achieving some balancing of exploration and exploitation and further enhancing this by embedding higher-level statistical distances in an improved fitness version, thereby providing EIGA the necessary tool for upholding statistical property for the datasets involved. Due of its genetic diversity, EIGA avoids time convergence, unlike many classic genetic algorithms. Iris, adults, and cardio benchmark datasets show that EIGA reduces RMSE and MAD best. RMSE improved from 0.1668 to 0.1654 and MAD from 0.0479 to 0.0455 on iris with 60% missing data. EIGA, however computationally expensive, is another good alternative for complicated datasets that need more precise imputation.
  Keywords: data intelligent imputation; dynamic crossover; dynamic mutation; genetic algorithm; fitness function; missing data; multivariate data; imputed data; root mean square error; RMSE; mean absolute deviation; MAD; enhanced imputation genetic algorithm; EIGA.
  DOI: 10.1504/IJIIDS.2025.10070762
   
  
• Optimisation enabled deep learning model for data privacy protection in blockchain networks using federated learning  
  by T. Premkumar, D.R. Krithika 
  Abstract: Recently, federated learning (FL) has been employed in blockchain networks to protect users data privacy. This paper proposes gradient beluga whale optimisation-deep residual network (GBWO-DRN) for data privacy protection in blockchain networks. Data privacy protection in the blockchain network is performed by nodes and servers. The process is performed in local and global training models, where distributed data is fed into a local model. The data is normalised and Laplace noise is added to it. Paillier homomorphic encryption is applied and the result is classified by the DRN. The GBWO is used to train DRN to improve DRNs performance. Finally, local updation and aggregation are done in the global training model and the data is stored in cloud. The GBWO-DRN recorded false positive rate (FPR), root mean squared error (RMSE), mean squared error (MSE), accuracy, loss function, and mean average precision (MAP) of 6.64%, 37.98%, 14.42%, 93.52%, 6.48%, and 92.10%.
  Keywords: deep residual network; gradient beluga whale optimisation; GBWO; gradient descent optimisation algorithms; GDOA; beluga whale optimisation; BWO; federated learning.
  DOI: 10.1504/IJIIDS.2025.10070801
   
  
• Honey badger optimisation based recurrent neural network for COVID-19 classification  
  by S. Salini, B. SelvaPriya 
  Abstract: A new research has begun to look into the sounds of people breathing, coughing, and recording their voices using hospital-confirmed COVID-19 tools. The sounds these things make are different from the sounds healthy people make. When COVID-19 was identified through coughing, the data on non-respiratory and respiratory noises that were linked to all the different situations that were given were also taken into account. The virus that causes the lung illness known as COVID-19 is called Severe Acute lung Syndrome Coronavirus-2 (SARS-CoV-2). The number of COVID-19 cases has been slowly rising, which more important to find safe ways to find people who have the virus. The Gaussian filtering model brings out the sensitivity of the method that was made better. The specificities of DCNN are 0.9125, BI-AT-GRU are 0.8926, and XGBoost are 0.9014. In k-fold value 9, a new DNFN based on JHBO has a precision of 0.9219.
  Keywords: honey badger optimisation; COVID-19 classification; feature extraction.
  DOI: 10.1504/IJIIDS.2025.10071128
   
  
• ECC based authentication protocol for IoT and untrusted cloud computing environments  
  by Sheetal Kalra, Jyoti Saroj 
  Abstract: Over recent years, password-based remote user authentication schemes using smart cards have become increasingly popular in internet of things (IoT) and cloud computing (CC) environment. However, implementing these smart card-based systems is costly due to the necessary infrastructure for smart card utilisation. In this research paper a dynamic password based remote user authentication protocol for common storage devices based on elliptic curve cryptography (ECC) has been proposed. This protocol not only preserves all merits of a smart card based authentication schemes rather reduces the computational cost as well. The comparison shows that the protocol is robust in comparison to other protocols and achieves all the security requirements. The security analysis confirms that the proposed protocol offers robust security and is impervious to network threats.
  Keywords: authentication; elliptic curve cryptography; ECC; smart card; mobile devices; internet of things; IoT; cloud computing; CC.
  DOI: 10.1504/IJIIDS.2025.10071134
   
  
• Quantum convolutional neural networks with new SVM-FE feature selection for robust intrusion detection  
  by Yogesh Gurav, Mukil Alagirisamy, Sathish Kumar Selvaperumal 
  Abstract: Intrusion detection is a vital component of cybersecurity, requiring effective methods to identify anomalies in complex datasets. Traditional machine learning models face challenges with high-dimensional data and intricate feature relationships. To address these, we propose an Improved quantum convolutional neural network (QCNN) combined with a novel SVM-RFE-based feature selection technique for enhanced detection. The process begins with data collection and pre-processing using min-max normalisation. Feature extraction captures various aspects of the dataset, including correntropy, statistical, and raw features. CASC-SVM-RFE is then applied for feature selection, reducing dimensionality while retaining key attributes. Anomaly detection is performed using the eigen value decomposition integrated convolutional layer within the QCNN, utilising quantum computing principles for accurate identification. Our method achieves 0.949 accuracy, outperforming traditional models like LSTM, Bi-LSTM, QNN, SVM, CNN, RNN, and QSVM+QCNN, showcasing the potential of quantum computing for tackling complex, high-dimensional intrusion detection tasks.
  Keywords: intrusion detection system; EDIC-QCNN; CASC-SVM-RFE; min-max normalisation and anomaly detection.
  DOI: 10.1504/IJIIDS.2025.10071348
   
  
• BioASQ-Ispec16: a novel resource for advancing specialty-based organisation in medical texts  
  by Walid Benaouda, Siham Ouamour, Halim Sayoud 
  Abstract: To improve specialty-specific classification of biomedical texts, we introduce BioASQ-Ispec16, a novel corpus of 844,481 medical abstracts spanning 16 specialties. This dataset undergoes rigorous filtering and cleaning to ensure high-quality annotations. We evaluate three main approaches: zero-shot learning, retrieval-based classification, and fine-tuned transformer models. Our results highlight the limitations of zero-shot learning, which, despite its flexibility, fails to capture domain-specific nuances, achieving an accuracy of 62.51% and an F1-score of 62.55%. In contrast, retrieval-based methods, particularly bge-large-en-v1.5, show improved performance, with an F1-score of 84.32%, demonstrating its effectiveness in medical abstract classification. However, fine-tuning BERT-like models on domain-specific data proves to be the most effective strategy, with accuracy ranging from 90.77% to 92.54%, and with PubMedBERT achieving an F1-score of 92.68%, outperforming other models. These results provide insights into enhancing medical text classification systems, highlighting the value of domain-specific data annotation and the effectiveness of transformer-based approaches.
  Keywords: medical text classification; retrieval-based models; bidirectional encoder representations from transformers; BERT; zero-shot learning; ZSL; large language models; LLMs.
  DOI: 10.1504/IJIIDS.2025.10071884
   
  
• An automated segmentation and classification model for leaf disease using the multiscale residual LSTM  
  by R.C. Dyana Priyatharsini, G. Rosline Nesa Kumari 
  Abstract: Plant leaf diseases are significant for reducing crop failure as well as the transferring of harmful viruses. In this work, we designed a unique plant disease classification model using deep learning models. Initially, the images are gathered from publicly available datasets. Then, the accumulated images are put into the segmentation stage done by the adaptive Bayesian clustering (ABC) model; here several parameters are tuned by the improved one-to-one-based optimiser (IOOBO) to enhance the effectuality of the designed approach. Then, the feature extraction procedure is conducted in the segmented images. The features like shape, colour, texture, and vision transformer (ViT)-aided features are extracted. Further, the multiscale residual long short-term memory (MR-LSTM) model is adopted for leaf disease classification with the help of extracted features and the classified images as the outcome. Finally, the experimental analysis of the recommended approach is executed and validated among various traditional models.
  Keywords: leaf disease segmentation; leaf disease classification; adaptive Bayesian clustering; ABC; multiscale residual long short-term memory; MR-LSTM; improved one-to-one-based optimiser; IOOBO.
  DOI: 10.1504/IJIIDS.2025.10072338
   
  
• Research on smart phone app interface interaction design based on smart city internet of things  
  by Xiao Chen, Yi Yu 
  Abstract: As the functions of mobile phone software become stronger and the corresponding industry development trends become more and more perfect, the design of mobile APP pages has also risen to the height of a special study. This article intends to conduct in-depth research on the human-computer interaction design of the smart phone APP interface through the smart city internet of things. This article first introduces the smart city under the internet of things. The QoS indicator of the internet of things is modelled, and the QoS calculation process based on the internet of things is given. The construction of smart cities was explored based on the internet of things environment, and the interactive design of smart phones was researched on the basis of the internet of things environment.
  Keywords: smart city; smart phone; interactive design; QoS indicator modelling.
  DOI: 10.1504/IJIIDS.2025.10072467
   
  
• Android malicious application detection technology based on deep learning  
  by Jing Liu, Mingwei Sun, Ying Meng 
  Abstract: With the widespread use of mobile phones, they have become essential tools in everyday life. Androids open architecture and commercial potential have made it a frequent target for malware and security threats. At present, the detection mode and technology of Android malware have a low detection rate and low efficiency and cannot effectively detect the currently rapidly growing malware. In the detection technology test, when using deep learning algorithms to detect malicious programmes, its recognition accuracy reaches 97.2%. Under the same dataset and input features, it is significantly better than the detection performance of traditional classifiers such as support vector machines, decision trees, naive Bayes, K-nearest neighbour algorithms, and mainstream deep learning models such as RNN, GRU and transformer. Experiments show that the detection effect of Bi-LSTM algorithm on malicious programmes is better than the above detection methods.
  Keywords: application detection technology; Android malicious application; deep learning; bidirectional long short-term memory neural network.
  DOI: 10.1504/IJIIDS.2025.10072577
   
  
• Parallel image filtering algorithm with adaptive dynamic load balancing  
  by Xiangjiao Liu, Rui Cui 
  Abstract: To solve the problem of waste of computing resources and low efficiency due to uneven load in the parallel image filtering process, this paper introduces an adaptive dynamic load balancing mechanism, combines the block parallel convolution filtering algorithm, dynamically adjusts task allocation, optimises the use of computing resources, and improves the filtering performance and real-time performance in large-scale image processing. A dynamic task allocation strategy is implemented in parallel image filtering to ensure that each processing unit can balance the load. A dynamic adjustment task scheduling strategy is introduced to balance the computing load of each processing unit, and the image is filtered using a block parallel filtering algorithm. The experimental results show that the adaptive load-balanced block parallel convolution filtering algorithm in this paper has an acceleration ratio, computing resource utilisation, memory efficiency and energy consumption ratio of 2.5, 85%, 220 MB/s and 1.35 J/s respectively, and is superior to other parallel image filtering algorithms of different models, demonstrating the high efficiency and real-time performance of this study in image filtering. This study presents a novel approach to mitigate waste of computational resources and inefficiencies in large-scale image processing.
  Keywords: adaptive dynamic; load balancing; parallel image; filtering algorithm; convolution filtering.
  DOI: 10.1504/IJIIDS.2025.10072999
   
  
• Data mining and pattern recognition in intelligent maintenance orders of power systems under knowledge graph  
  by Yin Wu, Yuechao Jin, Wuneng Ling, Jiayi Yang, Yan Qin, Fangling Luo 
  Abstract: The paper studied the entity relationships of power faults based on datasets and extracted entity relationships in the fault domain. Word2Vec can be used to mine textual data for maintenance orders and preprocess the mined data. Support vector machines can be used to construct fault recognition patterns for power systems, improving the accuracy of fault recognition for power system maintenance. It completed the generation of intelligent maintenance orders through the construction of recognition patterns and power knowledge graph. The accuracy rate, recall rate, and F1-values of the power system maintenance fault mode recognition model based on long short-term memory-support vector machine (LSTM-SVM) are 87.23%, 74.39%, 84.23%. The data mining and pattern recognition in intelligent maintenance orders of power systems based on knowledge graph can help improve the maintenance efficiency and resource allocation optimisation of power systems. It can also promote technological innovation and energy security, making positive contributions to the sustainable development of the power industry.
  Keywords: knowledge graph; power system maintenance orders; data mining; support vector machine; SVM; pattern recognition.
  DOI: 10.1504/IJIIDS.2025.10073064
   
  
• Corporate credit economy evaluation based on improved K-means clustering algorithm  
  by Zhifei Yi 
  Abstract: In todays dynamic economic environment, a firms credit status is crucial for financial stability, investment decisions, and business cooperation. This study designs a corporate creditworthiness evaluation method based on an improved K-means clustering algorithm. The process involves constructing an economic clustering index system, reducing dimensionality using factor analysis, and extracting key common factors. A binary classification approach is then applied. To enhance performance, a density- and weight-based K-means algorithm is proposed. Experimental results show that the proposed method outperforms two other algorithms in terms of convergence speed and optimisation accuracy. Specifically, on the vowel dataset, it converged 10.25% and 6.47% faster, and on the glass dataset, 8.67% and 7.45% faster than the other algorithms. These results demonstrate the efficiency and accuracy of the improved K-means algorithm in evaluating corporate credit.
  Keywords: K-means clustering algorithm; KCA; credit economy evaluation; binary classification research; high-density; factor loading matrix.
  DOI: 10.1504/IJIIDS.2025.10073065
   
  
• Electronic information rapid collection system based on artificial intelligence data intelligent algorithm  
  by Qiushi Song 
  Abstract: Given the problems of insufficient dynamic data collection efficiency and weak adaptability in the current electronic information rapid collection system, this paper applies an intelligent collection system architecture based on the improved deep reinforcement learning agent (DRL-Agent) model and adopts a software-hardware collaborative design, including a four-level architecture of data perception layer, hardware acceleration layer, edge computing layer, and hybrid intelligent decision-making layer. At the algorithm level, a three-dimensional feature tensor is innovatively constructed to fuse sensor time series data, device state metadata, and environmental parameters, and a dual-stream attention mechanism is designed to achieve joint modelling of spatiotemporal features. At the system level, a dynamic reward function is added to balance the delay, accuracy, and energy consumption objectives, and a cross-scenario knowledge transfer framework is combined to achieve adaptive learning of parameter freezing and fine-tuning. Finally, the effectiveness of the intelligent collection system is verified. The research results show that in dynamic scenarios such as intelligent manufacturing and smart cities, the system can reduce data collection delay by 87.5%, and the accuracy of most anomaly detection is over 90%. The improvement method adopted can simultaneously improve the dynamic data collection efficiency and adaptive capabilities.
  Keywords: DRL-Agent model; intelligent collection system; dual-stream attention mechanism; dynamic reward function; feature extraction.
  DOI: 10.1504/IJIIDS.2025.10073244
   
  
• A cutting-edge air quality monitoring and forecasting system with environmental protection and user accessibility: from prototype to reality  
  by Tan Duy Le, Binh Nguyen Le Nguyen, Kha Tu Huynh, Thanh Duc Nguyen, Nguyen Tan Viet Tuyen 
  Abstract: Air pollution is a significant problem, causing environmental and human health risks. To tackle this challenge, it is essential to design an efficient framework for monitoring and controlling air quality levels. Considerable efforts have been made to create efficient air pollution monitoring systems, but this domain still needs to overcome several obstacles. To address this, our research suggests the implementation of a hardware apparatus to gather data on meteorological conditions and pollution levels from the surrounding environment. This approach is complemented by a web and mobile application enabling users to monitor the collected data and make predictions for the next three hours. The designed system is further equipped with a high-performance microcontroller and integrated with several sensors for collecting environmental data including PM2.5, CO2, temperature, and humidity. The data collected is then sent to ThingSpeak, a cloud-based internet of things (IoT) platform, for data storage and API access. Our designed API enables advanced computation and visualisation on web and mobile applications for end-users to track air quality. We believe that our precise quantification and visualisation of atmospheric contaminants could support policymakers, researchers, and communities in facilitating well-informed decision-making and responding swiftly to issues related to air quality.
  Keywords: internet of things; IoT; air quality monitoring; cloud computing technology; web/mobile application; forecasting system.
  DOI: 10.1504/IJIIDS.2025.10073316
   
  
• Research on hot topic discovery in online medical communities based on density peak clustering algorithm  
  by Fan Tongke 
  Abstract: Difficulties in accessing medical care, high medical costs, and uneven distribution of medical resources are common problems. With the continuous development of social media and the widespread use of mobile smart devices, people have begun to accept online consultations as a form of medical care. As a result, patients demand for healthcare services has increased, and they are increasingly seeking precision treatment. This paper takes text data from patient consultations in online medical communities as its research object and proposes a text similarity measurement method called W2VTI based on the Word2Vec model, which incorporates semantic relationships and word weights. This method is used to identify hot topics in patient consultations in online medical communities, thereby providing theoretical and practical basis for the research directions of medical researchers and the service directions of medical professionals, and better meeting patients medical needs.
  Keywords: text mining; online medical community; text clustering; knowledge discovery.
  DOI: 10.1504/IJIIDS.2025.10073340
   
  
• A survey-based on present approaches and future research trends: predictive analysis on healthcare monitoring systems using deep learning techniques  
  by Vidhya Muthulakshmi Ramachandran, S. Malathi 
  Abstract: This survey explores the different methodologies for the predictive analysis of healthcare monitoring systems. This work follows the introduction, literature review, and chronological analysis of the healthcare monitoring systems model. This survey provides a detailed explanation of different sources of data collection for the predictive analysis of healthcare systems. It also explores the benefits of using different deep-learning techniques and algorithms for predictive analysis in the healthcare system. Deep learning techniques optimised for real-time data processing, enabling timely predictions that are essential for immediate clinical decision-making and treatments. Alongside this, the varied evaluation metrics are considered for illustrating the efficiency. Finally, the research gaps and challenging factors are given to direct the future development of predictive analysis on healthcare monitoring systems.
  Keywords: predictive analysis; healthcare monitoring systems; dataset details; disease types; predictive techniques; merits and demerits; performance index; research gaps and future directions.
  DOI: 10.1504/IJIIDS.2025.10073657
   
  
• Integration and security analysis of sensor technology and traditional encryption algorithms in 6G networks  
  by Zhenmei Yue, Yinna Bu 
  Abstract: To address 6G challenges like increased devices, complex environments, and quantum threats, this paper proposes a hybrid encryption scheme using quantum key distribution (QKD) and AES-128. BB84 generates quantum keys, combined with ECDH for secure transmission. Sensor encryption uses dynamic key refresh and hardware acceleration for efficiency. Distributed architecture and MQTT enhance transmission efficiency. Experiments show fibre-optic QKD outperforms free space, with a 100 kbps key rate at 5 km. In dense node scenarios, the key rate drops from 90 kbps to 55 kbps between 520 km, and error correction falls from 94% to 88%. The QKD+AES scheme achieves 99.99% confidentiality and 100% quantum resistance. Optimised AES consumes only 0.8W, making it ideal for resource-constrained environments, enhancing IoT security and adaptability in 6G networks.
  Keywords: 6G networks; quantum key distribution; sensor encryption; AES algorithm; MQTT protocol.
  DOI: 10.1504/IJIIDS.2025.10073666
   
  
• Mobile information system for spoken English learning based on wireless sensor network  
  by Xinxin Zhang, Lingzhi Xu, Lige Qiao, Jing Cai 
  Abstract: With the deepening of international communication, the importance of English-speaking ability in personal development, social and national foreign communication has become increasingly prominent. The traditional English-speaking learning system has problems such as large response delay and single learning module, which makes it difficult to meet the needs of the times. This paper innovatively integrates wireless sensor networks into the design of mobile English-speaking learning information system, and proposes improved intelligent clustering and learning automaton algorithms. The improved ICLA algorithm optimises cluster head selection by comprehensively considering node residual energy and regional node density, and adjusts the number of members in the cluster to achieve load balancing. This improvement improves the rationality of network node clustering, speeds up response speed and reduces energy consumption. Experimental data show that the response speed of the ICLA algorithm before the improvement is between 53.8% and 59.9%, and it increases first and then decreases with the increase of the number of network nodes.
  Keywords: international exchange; English oral learning; wireless sensor network; WSN; ICLA algorithm; mobile information system.
  DOI: 10.1504/IJIIDS.2025.10074117
   
  
• A statistical analysis of deepfake text on social-media using enhanced RoBERTa  
  by Adamya Gaur, Sanjay Kumar Singh, Pranshu Saxena 
  Abstract: Recently, there has been rapid development in natural language understanding and generation, resulting in machines enhancing coherent text generation ability. Vicious people may utilise the technology to generate false synthetic text to navigate peoples views and opinions. To tackle the alarming situation, the study focuses on detecting deepfake text as fake tweets on the social media platform X (formerly Twitter), utilising various deep learning architectures and GloVe and FastText embeddings. A hybrid model is also proposed using the RoBERTa model, which is specially trained and fine-tuned for tweets, and a bi-directional gated recurrent unit (GRU). This study starts with data processing modules that recognise stop words, missing values, punctuations, URLs, categorical columns, tokenisation, and padding. After pre-processing, feature engineering selects and optimises important feature-space vectors. These characteristic vectors are fed into several classification methods to compare proposed and conventional deep networks. The proposed model shows 91.7% accuracy, which outperforms state-of-the-art architecture.
  Keywords: deepfake text; social media; fake tweet; machine learning; language processing.
  DOI: 10.1504/IJIIDS.2025.10074752
   
  
• Discrete spotted hyena optimiser for extracting multiple-choice tests  
  by Tram Nguyen, Van Du Nguyen 
  Abstract: Multiple-choice test extraction is a multi-constraint optimisation problem, requiring a balance of several factors (i.e., question difficulty, diversity, and minimising overlap between tests). This paper proposes a novel approach to multiple-choice test generation called the discrete spotted Hyena optimiser (D-SHO). To enhance exploration and avoid local optima in difficulty-level optimisation, this study adopts two local search strategies swap operator (SO) and swap sequence (SS) which also increase the initial population diversity in D-SHO. Additionally, the paper introduces domain-specific operations for multi-objective test generation scenarios involving difficulty level and time requirements, namely PD-Shift and PD-Exchange, which are applied to strengthen the local search, phase effectively and ensure fairness in generating multiple tests with a consistent difficulty level for a group of students. Experiments conducted on a question bank with 1,000 and 12,000 questions indicated the proposed methods scalability, robustness, and performance.
  Keywords: question bank; spotted hyena optimiser; multiple-choice tests; single-objective optimisation; multi-objective optimisation.
  DOI: 10.1504/IJIIDS.2025.10074933
   
  
• Cloud computing data network security monitoring system based on chaotic cryptography and RSA algorithm  
  by Yang Gao, Xuezhong Lu, Gulixiati Abulikenmu, Bota Muheyat 
  Abstract: Cloud computing (CC) data network security monitoring systems face the problem of key management and encryption algorithm vulnerability. Chaotic cryptography improves encryption strength by generating complex keys, while the RSA algorithm enhances the security of key exchange. This paper combines the two approaches and builds a hybrid encryption model to address the security vulnerabilities of the existing system, ensuring the security and integrity of data transmission. Experiments show that the model has a 10% success rate for brute force cracking and a 1.2% success rate for man-in-the-middle attacks. The encryption and decryption times are optimised to 3.0 ms and 2.5 ms. The key generation rate is stable at 5870 keys/sec, and the ciphertext information entropy reaches 5.46.2 bits, which has both high security and reasonable computing overhead. The results show that the scheme effectively balances encryption performance and anti-attack capabilities, solves vulnerabilities in key generation, transmission protection, and quantum threats, and provides a reliable encryption framework that considers efficiency and security for CC and big data scenarios.
  Keywords: monitoring system; data network security; cloud computing; chaotic cryptography; RSA algorithm.
  DOI: 10.1504/IJIIDS.2025.10075130
   
  
• An efficient automatic cancer classification using Haliaeetus unicinctus optimisation-based deep LSTM  
  by Swapnil Ashokrao Bobade, Suresh S. Asole 
  Abstract: Automated cancer classification is crucial, yet training-related issues often hinder success. However, only a few feature genes are strongly associated with cancerous lesions in the high-dimensional datasets. Consequently, it is critical to precisely choose a subset of feature for accurate cancer classification. Hence, this proposed research introduces the Haliaeetus unicinctus optimisation-based deep long short-term memory (HUO-deep LSTM) method for automated cancer classification using microarray gene expression data. Utilising the HUO algorithm into the feature extraction, permits the proposed approach to identify the optimal subset of features. The proposed approach extracts the discriminative features and enhances the later application of LSTM networks that capture the long context correlations in a microarray dataset. Notably, the proposed approach achieves the high accuracy of 97.50%, sensitivity of 97.87%, and specificity of 95.51% for microarray gene expression dataset. For the gene expression prediction dataset, metrics are 98.46%, 96.88%, and 97.73%, surpassing the other existing methods.
  Keywords: cancer classification; deep long short term memory; Haliaeetus unicinctus optimisation; HUO; bald eagle optimisation; Harris hawk algorithm.
  DOI: 10.1504/IJIIDS.2026.10075600
   
  
• Improving online text-based person search by incorporating gender and age information  
  by Hoang-Son Bui, Thi-Hoai Phan, Thi-Ngoc-Diep Do, Hung-Phong Tran, Thanh-Hai Tran, Thi-Lan Le 
  Abstract: Text-based person search has recently gained attention due to its potential applications. Previous studies have achieved impressive results in offline evaluations using within-domain test sets. However, when deploying the application for end users, the search performance significantly decreases due to the diversity of queries. This paper aims to bridge the gap between offline and online performance. We introduce an online framework for text-based person search in Vietnamese in which a new module is added in the text-based person search system that refines search results based on gender and age group estimation. Two strategies are proposed to re-rank the retrieved list from gender and age group information. Experimental results show that the proposed model leads to improvements in retrieval precision of 3.23%, 2.53% and 27.19% when using age estimation, gender-age estimation and gender estimation with the strategy 1 and of 5.69%, 7.03%, and 27.69% with the strategy 2, respectively.
  Keywords: text-based person search; online person search; person search in natural languages; gender and age estimation.
  DOI: 10.1504/IJIIDS.2026.10075601
   
  
• Feasibility study of online English reading course based on data mining  
  by Wenna Xu, Li Xiong, Ying Zhang, Yu Sun 
  Abstract: In the era of information explosion, extracting meaningful insights from massive data and presenting them through visual charts is essential for effective decision-making. Current online English reading courses lack personalised content matching and efficient course selection mechanisms, leading to suboptimal user experience. This study addresses this gap by applying data mining techniques to optimise course recommendations. The proposed solution utilises association rule, classification, and clustering algorithms to analyse user behaviour and course preferences. Experimental evaluation on a dataset comprising online course interactions demonstrates a 15% improvement in course selection efficiency. The findings confirm the feasibility of data mining in enhancing the accuracy and personalisation of online English reading courses, providing a scientific basis for course recommendations.
  Keywords: online English reading course; data mining; association rule algorithm; classification algorithm.
  DOI: 10.1504/IJIIDS.2026.10076002
   
  
• Multi-channel speech recognition enhancement system based on neural network algorithm  
  by Guoqiang You 
  Abstract: Even with advances in close-talk sound identification, distinguishing distant speakers remains difficult due to significant reverberation and overlapping speech. This research proposes beamforming and speech separation networks for automatic speech recognition using neural networks (ASR-NN) to address these issues. In this paper, ASR-NN offers a simple yet effective technique for recognising multi-channel, far-field, overlapping speech. A convolutional neural network (CNN) and multi-channel first-order ambisonics mixes are used to enhance speech. Automated speech recognition systems using data-dependent spatial filters built using a mask-based technique can handle reverberation and competing speakers. Another competing speaker is added to the long short-term memory (LSTM) network and tested under more stressful conditions than those used in prior research. The new system includes bidirectional LSTM (BLSTM) mask estimation based on beamforming, a BLSTM-CNN hybrid acoustic model, and LSTM language model rescoring. This is accomplished by using an adaptive speaker system that utilises limited re-estimation techniques in conjunction with several acoustic models and beamforming strategies. The ASR-NN technique uses word-posterior system combination techniques to combine these systems.
  Keywords: long short-term memory; LSTM; convolutional neural network; CNN; automatic speech recognition; neural networks.
  DOI: 10.1504/IJIIDS.2026.10076177
   
  
• Elucidating the intelligent model of fake news detection using dilated and attention-based Bayesian LSTM with weighted feature integration and improved heuristic approach  
  by Ginika Mahajan, Deepika Shekhawat, Anita Shrotriya, Harish Sharma, Anju Kalwar, Harsh Taneja 
  Abstract: Fake information might sway public opinion and spread quickly because of todays information-rich digital environment. In this work, a novel fake news recognition model using a heuristic-aided algorithm is suggested. Initially, text data is congregated from appropriate datasets. The raw texts go through the pre-processing to eliminate unnecessary data. The pre-processed text data is given as input to the term frequency-inverse document frequency (TF-IDF), bidirectional encoder representations from transformers (BERT), and text convolutional neural network (TextCNN) approach helps to recover the significant features. From the resultant of extracted features, the weighted feature fusion is acquired, in which the weights are optimally tuned by recommending the novel algorithm fitness aided garter snake optimisation (FAGSO). At last, fake news detection is accomplished by the dilated and attention-based Bayesian long short-term memory (D-AtBLSTM). The implemented fake news detection performance is contrasted to traditional fake news detection techniques, and it displayed high accuracy.
  Keywords: fake news detection; term frequency-inverse document frequency; bidirectional encoder representations from transformers; BERT; text convolutional neural network; TextCNN; fitness aided garter snake optimisation; FAGSO; dilated layer; attention mechanism; Bayesian long short-term memory.
  DOI: 10.1504/IJIIDS.2026.10076285
   
  
• Improved simplified fractional Fourier transform with hybrid black widow-elephant herding optimisation for detecting moving objects on ground  
  by Talla Neelima, Tirumala Krishna Battula 
  Abstract: Ground moving target indication (GMTI) is effectively accomplished with the help of synthetic aperture radar (SAR), which includes the distributed moving and point moving objects. Displaced phase centre antenna (DPCA) is the GMTI technique that is currently used. The detection process is done by utilising the objects speed and signal-to-clutter ratio (SCR). The GMTI between the background clutter and moving objects is calculated. But, the detection process is not accurate because of low SCR caused via the specular scattering. Hence, this paper plans to implement a new Improved simplified fractional Fourier transform (ISFrFT) based on the hybrid meta-heuristic algorithm. This proposed method is used for estimating the moving targets Doppler parameters. The hybrid black widow elephant herding optimisation (HBWEHO) is used for improving the performance of the target estimation with developed ISFrFT. The performance of the developed model was evaluated against various existing approaches and showed that the computational time of the developed HBWEHO-ISFrFT is 0.0295 sec, which is lesser than PSO-ISFrFT, EFO-ISFrFT, BWO-ISFrFT and EHO-ISFrFT. The experimental analysis ensures the promising performance of the developed target estimation method towards the moving target.
  Keywords: GMTI; simplified fractional Fourier transform; black widow optimisation; elephant herding optimisation; synthetic aperture radar; SAR.
  DOI: 10.1504/IJIIDS.2026.10076388
   
  
• Deep learning-based image retrieval and classification model with multi-similarity using transformer enabled multi-scale Yolov5 with attention mechanism  
  by Davuluri Rajya Lakshmi, Dhupam Bhanu Mahesh, Cheekati Bindu Madhuri 
  Abstract: The images are collected from the publically available data source, and it is given to the segmentation stage during training phase. In this segmentation stage, a transformer-enabled multi-scale Yolov5 with attention mechanism (TM-YOLO5-AM) is utilised for segmenting the images. Then, the segmented images are fed into the atrous spatial pyramid pooling-based residual attention network (ASPP-RAN) for classification. Furthermore, the classified images are stored in the database. In the testing phase, the user-related query images are taken for retrieving the appropriate images from the database. Here, the deep features from the query images are extracted using an ASPP-based RAN approach. These extracted features are compared with the stored and trained image features in the database by using the multi-similarity function. Here, several similarity measures are applied to check the multi-similarity, and based on the similarity measures, the appropriate images will be retrieved within a short time.
  Keywords: image retrieval and classification; transformer-enabled multi-scale Yolov5 with attention mechanism; atrous spatial spectral polling-based residual attention network; similarity checking.
  DOI: 10.1504/IJIIDS.2026.10076588
   
  
• A GWO-based approach to university exam timetabling problem: a case study  
  by Van Du Nguyen, Tram Nguyen 
  Abstract: Swarm intelligence (SI) has proven to be an effective approach for solving optimisation problems. Among SI techniques, the grey wolf optimiser (GWO), inspired by the hunting behaviour of grey wolves, has attracted attention due to its impressive features over others. This paper addresses the university exam timetabling problem, a complex and practical scheduling task involving the assignment of exams to limited timeslots and rooms under predefined constraints. However, each educational institution has its constraints in the scheduling process. We propose a discrete GWO-based approach tailored to this problem. Experimental results demonstrate that the proposed method outperforms simulated annealing (SA) and genetic algorithm (GA) in terms of timetable quality and computational efficiency. In addition, hybrid approaches combining SA with GA and GWO are evaluated to further analyse performance.
  Keywords: metaheuristics; swarm intelligence; exam timetabling; collective intelligence.
  DOI: 10.1504/IJIIDS.2026.10076589
   
  
• DELEA-XLNet: double exponential lotus effect optimisation algorithm-based XLNet for consumer sentiment analysis  
  by N. John Kuotsu 
  Abstract: The main approach to enhancing the development of online reviews is through sentiment analysis which is an attractive field within both academic research and industries. The review serves several sectors, but collecting accurately interpreted training data remains challenging. Thus, this work introduces an efficient method for analysing consumer sentiment. At first, the preprocessing is done by stemming and stop word removal. Then pre-processed data is fed to feature extraction, while the SentiNet features are extracted using XLNet and then the feature selection is executed by wrapper approach. At last, sentiment analysis is done using 1-dimensional convolutional neural network (1DCNN), and it is trained using proposed double exponential lotus effect optimisation algorithm (DELEA), which is developed by the combination of DES and LEA. At last, experimentation analysis exhibits that DELEA model achieved improved accuracy, sensitivity, and specificity with values of 0.934, 0.950 and 0.904.
  Keywords: sentimental analysis; review text; rating prediction; deep learning; optimisation algorithm.
  DOI: 10.1504/IJIIDS.2026.10076716
   
  
• Communication network security intrusion detection system based on data intelligence algorithm  
  by Yuansheng Du 
  Abstract: In order to solve the contradiction between high-dimensional and unbalanced data processing and real-time detection efficiency in data intelligent algorithms for communication network security intrusion detection systems, this paper proposes a detection system framework that integrates attention mechanisms and lightweight models. Meanwhile, this paper proposes a sliding window dynamic threshold strategy, which effectively reduces the false alarm rate. This experiment was validated using large-scale communication network traffic data. The results show that the false alarm rate of the system in a typical network environment is 1.2%, the response time is 8.34ms, and the average F1 score for multiple types of attacks is 90%. Lightweight design not only optimises the resource consumption of the model, but also improves the system's adaptability to real-time data streams. The dynamic threshold strategy has demonstrated robustness in constantly changing network environments and effectively responded to fluctuations in traffic and attack patterns in communication networks.
  Keywords: data intelligence algorithm; communication network security; intrusion detection system; attention mechanism; lightweight model.
  DOI: 10.1504/IJIIDS.2026.10077184
   
  
• Modelling of a bulk fuzzy queue system with two parameters and three parameters by trapezoidal fuzzy numbers using -cut methods  
  by G. Somasundara Ori, B. Abirami 
  Abstract: We study a model of the batch arrival fuzzy queuing system in this paper. The rates of service and arrival are both ambiguous. Trapezoidal fuzzy numbers were utilised in the computation of the algorithms efficiency metrics. The fuzzy queue measurements were transformed into crisp metrics using the Zadeh extension and the -cut method. Since various clients or consumers are supplied by various types of processors that comply with established queue control, queueing frameworks are utilised regularly in real-world scenarios. The calculation of the performance measures for priorities two and three with equal service rates and varied service rates has been illustrated with a numerical example.
  Keywords: bulk queuing model; priority queues; parametric programming problem; fuzzy trapezoidal fuzzy number.
  DOI: 10.1504/IJIIDS.2026.10077246
   
  
• An effective framework to perform lung nodule segmentation using U-net network  
  by Kha Tu Huynh, Huu Sy Le, Tân Le-Duy, An Mai 
  Abstract: Our paper proposes an approach to enhance the nodules in the lungs using a U-Net network with pre-processed data. The approach utilises a convolutional neural network (CNN) based architecture, namely U-Net, to accurately segment nodules from the surrounding lung tissue. The data pre-processing is done by applying a K-means algorithm and a median filter to the images, which helps reduce the noise and enhance the details of the no when segmenting. The U-Net network is then trained using the pre-processed images. The performance of the U-Net network is evaluated using a dataset of CT scans of the lungs. The results show that our model gain 0.836 with the loss error about 0.015 in the dice metrics, which is a promise result compared to the traditional U-Net network and CNN-approach techniques. We also embed the convolutional long short-term memory (Conv-LSTM) in our network to gain more accuracy due to the effectiveness of this network when extract more detail semantic information such as edges, shapes, . within fewer and light parameters. The proposed approach is effective for segment nodules from the lungs, which can be used in the diagnosis and treatment of lung diseases.
  Keywords: image segmentation; image enhancement; U-Net network; convolutional neural network; CNN; lung segmentation; K-means.
  DOI: 10.1504/IJIIDS.2026.10077398
   
  
• Distribution estimation algorithm for Bayesian statistical inference of multi-source heterogeneous data  
  by Ying Xu 
  Abstract: This paper proposes a multi-source adaptive distribution estimation algorithm based on a hierarchical Bayesian variational inference framework to address the problems of low inference efficiency and large posterior estimation bias caused by the failure of cross source heterogeneous fusion in traditional Bayesian distribution estimation algorithms in multi-source heterogeneous data scenarios. Firstly, this paper maps heterogeneous data to a shared latent space through a probabilistic embedding layer to achieve feature alignment; secondly, constructs a hierarchical Bayesian model to model cross source dependencies through global parameter priors and shared latent variables; then, adopts a split aggregation strategy to achieve distributed variational inference; finally, applies multi-source consistency loss to regularise the deviation between local posterior and global distributions. The experimental results show that when the conflict intensity increases from 10% to 50%, the post Kullback-Leibler divergence (KL divergence) of this method only increases to 0.073.
  Keywords: multi-source data; Bayesian inference; variational framework; posterior estimation; Kullback-Leibler divergence; KL divergence.
  DOI: 10.1504/IJIIDS.2026.10077531
   
  
• Data security transmission methods based on blockchain technology  
  by Wanpeng Yang, Ahmat Ablimit, Xuefei Su, Alai Tuerding 
  Abstract: To address the centralised risks and security challenges associated with data tampering in traditional data transmission methods, this study explores the use of blockchain technology to achieve more secure and efficient data transmission. By applying the distributed architecture of blockchain to decentralise data storage and verification processes, the risk of centralisation is effectively avoided. The consensus mechanism, applied on a blockchain, ensures that data remains complete and consistent during transmission. Encryption technology and digital signatures are used within this framework to enhance the security and reliability of data transmission. To further enhance security, real-time anomaly detection and machine learning algorithms are combined during data transmission. A secure cross-chain data transmission mechanism is also introduced to ensure consistency and atomicity between heterogeneous blockchain networks, supported by two-phase commit protocols, dual-signature validation, and zero-knowledge proof-enhanced key exchange. The findings demonstrate that this method can improve the security and integrity of data transmission and optimise the transmission efficiency to a certain extent. The anti-attack capability under the decentralised architecture scores five points, and the lowest integrity verification rate reaches 97.4%. Research has shown that combining blockchain with machine learning provides a new and efficient solution for secure data transmission.
  Keywords: blockchain technology; data security transmission; machine learning; support vector machine; SVM; consensus mechanism.
  DOI: 10.1504/IJIIDS.2026.10077600
   
  
• Social network relationship mining and visualisation analysis based on graph database  
  by Xinrui Xia, Bo Wang 
  Abstract: This paper proposes an improved DyGFormer model that integrates a Neo4j graph database and multidimensional visualisation to address challenges in dynamic social network analysis. It designs a dynamic attribute graph model with microsecond time granularity, achieving real-time updates and optimised query efficiency. An improved Louvain algorithm with a time-decay factor and Node2Vec verification improves community division consistency, while edge betweenness and community evolution detection enable cross-scale modelling. Visualisation, using D3.js and Neo4j Bolt, renders topology evolution and tracks propagation paths via A*. Experiments show that normalised mutual information in dynamic community detection improved from 0.72 to 0.76 over ten weeks, and real-time query response time for millions of nodes dropped to 2.3 seconds, significantly enhancing dynamic processing capabilities.
  Keywords: dynamic social network analysis; improved DyGFormer model; Neo4j graph database; dynamic community detection; visualisation interaction.
  DOI: 10.1504/IJIIDS.2026.10077691
   
  
• White blood cell subtypes classification using capsule networks for automated hematological diagnostic  
  by Shaili Gupta, Sanjeev Thakur, Deepti Mehrotra 
  Abstract: The accurate classification of white blood cells is essential for diagnosing and monitoring hematological disorders. This study presents a novel classification framework that integrates advanced preprocessing, balanced resampling, feature engineering, and deep learning to reliably categorise WBCs into five subtypes: neutrophils, lymphocytes, monocytes, eosinophils, and basophils. Unlike existing approaches that often rely on multi-focus imaging or computationally intensive transfer learning, the proposed method emphasises robustness and efficiency. Preprocessing techniques such as histogram equalisation and RandStainNA colour normalisation reduce staining variability, while synthetic minority oversampling technique (SMOTE) combined with class weight adjustment effectively addresses class imbalance. A multi-level feature extraction pipeline incorporating morphological, textural, and CNN-based features captures both global and localised cellular characteristics. For classification, a capsule network (CapsNet) is employed, achieving an accuracy of 97.38% under 5-fold cross-validation, thereby outperforming conventional models such as FCNN, SVM, random forest, and XGBoost. The novelty of this work lies in the seamless integration of class balancing and colour normalisation with CapsNet-based hierarchical feature learning, enabling superior generalisation with a streamlined workflow. The results establish the proposed framework as a competitive and clinically relevant approach, demonstrating its potential to advance automated white blood classification for reliable hematological diagnostics.
  Keywords: white blood cells; WBC; classification; machine-learning; image analysis; feature extraction; deep learning; medical diagnostics.
  DOI: 10.1504/IJIIDS.2026.10077692
   
  
• A novel improved deep score recurrent convolutional neural network-based tomato plant disease identification framework  
  by Sowmiya Murkesh, Krishnaveni Sivamohan 
  Abstract: To improve the agricultural market in India, early testing of plant leaves was very important. This study examines different features from images to identify diseases in tomato plants. Information was gathered from images of both healthy and diseased tomato plants for classification. Colour statistics, Hu moments, Haralick features and local binary patterns (LBP) were used alongside machine learning for training and testing models. An enhanced crossover bee optimisation (ECBO) method was applied to simplify the process and enhance performance. In this research, ECBO helped optimise two models: Visual Geometry Group (VGG16) and Inception-V3. After completing the image segmentation, the focus shifted to extracting features. An improved deep score recurrent convolutional neural network (IDS-RCNN) with the TensorFlow application programming interface (API) was utilised to classify crop diseases and detect them based on various characteristics.
  Keywords: plant leaf detection; machine learning; deep learning; classification; segmentation process.
  DOI: 10.1504/IJIIDS.2026.10077768
   
  
• English digital transformation algorithm for distributed big data based on Spark  
  by Xiaochao Yao 
  Abstract: In the era of big data, the main challenge of digital transformation lies in the inability of traditional text clustering algorithms to efficiently process large-scale English data, resulting in low adaptability and delayed information extraction that hinder intelligent decision-making. This study was conducted to enhance the efficiency and interpretability of digital transformation in English text analysis. A distributed framework based on Apache Spark and an improved K-means algorithm is proposed to overcome the scalability and accuracy limitations of traditional methods. This method combines the density peak and maximum and minimum criteria, and significantly improves the efficiency and accuracy of clustering by accurately selecting the initial clustering centre and optimising the calculation process. Experimental results show that the improved algorithm has a clustering accuracy of 10.53% higher than that of the traditional algorithm on multiple datasets, and shows higher stability and performance when processing large-scale data. In summary, the English digital transformation algorithm based on Spark shows superior performance than traditional methods in a big data environment and has strong practical value.
  Keywords: big data; distributed computing; Apache Spark; digital transformation algorithms in English.
  DOI: 10.1504/IJIIDS.2026.10077800
   
  
• Firefly optimisation algorithm for distributed network security vulnerability scanning and intrusion detection  
  by Cong Li 
  Abstract: To overcome challenges with blind spots and low detection accuracy, this study proposes a firefly optimisation algorithm for vulnerability scanning and intrusion detection in distributed network security systems. The author designed a distributed architecture and functional modules in accordance with the requirements, discussed the implementation process of each function based on the design, and tested the system performance parameters. The base classifiers were constructed by increasing the differences between the samples of each classifier, both by resampling the dataset and by determining the feature set; the detection results were integrated through weighting using varying learning algorithms to enhance diversification. In creating the weights, the firefly optimisation algorithm was employed to optimise the weighting of each base classifier. Experimental results indicate that the proposed algorithm performs better than other detection methods, as it produces a curve closer to the upperleft corner of the ROC plot. These findings further support the benefits and efficacy of the proposed approach. The method can consistently maintain a high level of detection accuracy (with a minimum of 96.6%) compared to other techniques.
  Keywords: machine learning; firefly optimisation; heterogeneous integration; and intrusion detection; vulnerability scan.
  DOI: 10.1504/IJIIDS.2026.10077847
   
  
• Enhancing dietary and herbal supplement planning through mBioBERT: a domain-specific approach to personalised medicine  
  by Pranshu Saxena, Vikas Tyagi, Mandeep Singh, Sanjay Kumar Singh 
  Abstract: This study investigates BioBERT for dietary and herbal supplement planning and introduces mBioBERT, a domain-specialised adaptation trained on curated dietary/herbal corpora and regulatory sources. The framework supports key biomedical NLP tasks (named-entity recognition, relation extraction and extractive QA) to capture supplement terminology and herbdrug/condition interactions. Compared with general BERT, conventional BioBERT and GPT-4 baselines, mBioBERT achieves 94% accuracy, 91% precision and 90% recall, indicating superior handling of complex relationships and domain-specific language. We describe design choices (entity-aware masking, sparse/global attention and cross-task distillation), report ablations, and discuss deployment considerations (latency, privacy and auditability). Limitations include dependence on high-quality domain data and computation for pretraining/fine-tuning, as well as degradation under heavy noise without denoising. Future work will expand datasets to broader supplement classes and conditions, incorporate real-world clinical text for robustness, and explore multimodal and multilingual extensions. Overall, the findings suggest mBioBERT can enable safer, more accurate and personalised supplement recommendations when used as clinician-supervised decision support.
  Keywords: dietary planning; herbs treatment; natural language processing; medicine; BioBERT; deep learning.
  DOI: 10.1504/IJIIDS.2026.10078004
   
  
• Bi-level optimisation for integrated production and logistics batch scheduling in discrete manufacturing  
  by Weitao Wei 
  Abstract: This paper proposes a method of integrating both production and logistics batch scheduling by developing an optimised twolayered system. The top layer optimises production batching using an improved mixed integer programming (MIP) model, while the lower layer dynamically schedules logistics batching with multiple optimisation objectives. Nondominated Sorting Genetic Algorithm II (NSGA-II) serves as the foundation for creating a bilevel scheduling optimisation model (BLSOM). This approach is an effective method for determining the entire Pareto front (the solution set) in a complex setting, thereby enabling overall optimisation of both production and logistics batches. Experimental results demonstrated that while the MIP model can find optimal solutions for smallscale instances, it becomes computationally prohibitive as the scale of the problem increases. In contrast, the proposed NSGAIIbased BLSOM exhibited superior performance in largescale practical scenarios, identifying nearoptimal solutions with substantially higher computational efficiency. Compared with the solutions obtained by the MIP solver within a limited timeframe, the proposed method yielded lower average costs (18.5%), reduced average delivery times (18.8%), and increased average utilisation of logistics resources (18.4%).
  Keywords: integrated scheduling optimisation; discrete manufacturing enterprises; production and logistics coordination; multi-objective genetic algorithm; resource utilisation efficiency.
  DOI: 10.1504/IJIIDS.2026.10078066
   
  
• NoSQL2SQL: a NoSQL to SQL database conversion tool  
  by Ranjeetsingh Suryawanshi, Alesha Mulla, Abhishek Saraf, Arya Chavan, Ritik Arora 
  Abstract: With the rise of big data, NoSQL databases have been used extensively because of their flexibility in handling unstructured as well as semi-structured data. However, data analysis, reporting, and most business intelligence tools require databases to have a defined, rigid schema. This idea presents NoSQL2SQL, an automated database migration framework that takes NoSQL databases like Redis, CouchDB, Cassandra, and Neo4j, and creates relational MySQL schemas with minimal data loss. It makes use of statistical schema analysis to infer optimal relational structures from document-oriented data, automatically detects relationships, infers data types for attributes and handles nested structures and arrays. A simplistic graphical user interface allows for ease of use. Experimental results show that the tool is successful in producing normalised relational schemas from NoSQL database inputs, making it suitable for analytics while ensuring that semantic relationships in the original data are not disturbed.
  Keywords: NoSQL; structured query language; SQL; database migration; schema inference; data integration.
  DOI: 10.1504/IJIIDS.2026.10078161
   
  
• CHKDE-SMOTE: leveraging convex hull and kernel density estimation approach for class imbalance in medical datasets  
  by Kaikashan I. Siddavatam, Subhash K. Shinde 
  Abstract: Addressing class imbalance is essential in machine or deep learning, especially in medical applications, as it leads to inconsistent and biased results. The adverse impact further worsens due to the intrinsic characteristics of the data, such as noise, class overlaps, and outliers. Although several imbalance handling techniques exist, outliers caused by human and measurement errors remain difficult to manage and can cause wrong interpolation of data. Existing SMOTE variants rely on local distance metrics and clustering assumptions, limiting their effectiveness for irregular data. To overcome these limitations, we propose convex hull and kernel density estimation-based SMOTE (CHKDE-SMOTE), a novel imbalance handling method that integrates global geometric structure and density validation, making it suitable for medical datasets with irregular distributions. The method employs a non-parametric convex hull depth strategy to remove boundary and outlier minority samples before synthesis and incorporates KDE-based sanitisation to generate reliable synthetic minority instances. The effectiveness of CHKDE-SMOTE is evaluated against multiple imbalance handling techniques using support vector machine, XGBoost, and random forest classifiers. CHKDE achieves an average F1-score improvement of 25.68% across classifiers and 39.30% on highly imbalanced datasets. Statistical significance is confirmed using Friedman test followed by Holm post-hoc analysis.
  Keywords: convex hull; imbalance data; kernel density estimation; SMOTE.
  DOI: 10.1504/IJIIDS.2026.10078201
   
  
• Real-time precise 4D trajectory prediction in complex low-altitude scenarios using multi-sensor fusion  
  by Chenchuan Zhang, Zhenkai Zhang, Ni Chen 
  Abstract: In order to solve the difficulties of multi-sensor data fusion and real-time accurate trajectory prediction in low altitude flight scenarios, this paper proposes a four-dimensional trajectory prediction method based on convolutional neural networks and transformer models. This method predicts the three-dimensional spatial position and temporal changes of the aircraft at future time steps. Firstly, this paper utilises convolutional neural networks to extract features from spatial data collected by various sensors such as vision, radar, and LiDAR. Secondly, this paper uses a transformer to model time series data. In addition, this paper proposes an innovative multi-level attention mechanism to optimise the multi-sensor data fusion process. Finally, the model can better evaluate the importance of different sensor data in complex environments. The results indicate that the method achieved a root mean square error of 0.82 metres in position with a prediction time step of 1.0 second.
  Keywords: low-altitude flight scenarios; multi-sensor data fusion; four-dimensional trajectory prediction; convolutional neural network; transformer model; multilevel attention mechanism.
  DOI: 10.1504/IJIIDS.2026.10078272
   
  
• A multi-relational graph convolutional network for multimodal sentiment analysis  
  by Huy-Minh Dau, Huyen Trang Phan 
  Abstract: Multimodal sentiment analysis (MSA) has critical applications in fields from human-computer interaction to mental health monitoring. Methods based on graph convolutional networks (GCNs) have been very effective thanks to their ability to model structural relationships. However, current GCN-based methods often treat these relationships in the same way. To address this limitation, our study proposes a novel architecture, which combines a relational graph convolutional network (RGCN) for text and a standard GCN for visuals. The core of the method is the construction of a text graph that simultaneously encodes both syntactic and semantic information. Information from the two branches is then combined through the attention mechanism. Thorough experimentation and a detailed ablation study on the CMU-MOSI dataset demonstrated the superiority of the proposed method. The full RGCN model achieved the best results compared to simpler GCN versions, affirming the value of explicitly modelling diverse relationship types.
  Keywords: multimodal sentiment analysis; MSA; emotion recognition; graph convolutional network; GCN; relational graph convolutional network; RGCN; multimodal fusion; attention mechanism; affective computing.
  DOI: 10.1504/IJIIDS.2026.10078616
   
  
• Personal information anonymisation processing technology and application based on big data  
  by Zhaoji Chen 
  Abstract: In big data environments, traditional anonymisation techniques for personal information face challenges with insufficient privacy protection and significantly reduced data utility due to data size and complexity. This paper proposes a local differential privacy and synthesis fusion model (LDPSyn). This model injects formalised privacy-preserving noise by applying an optimised random response perturbation mechanism to the categorical attributes of personal information at the data collection end. The perturbed data is then trained into a generative adversarial network based on the Wasserstein distance to learn its statistical distribution. Ultimately, a synthetic dataset is generated that conforms to the global characteristics and association rules of the original personal information, completely mitigating the risk of record-level re-identification through LDP perturbation and Wasserstein generative adversarial network (WGAN)-based synthesis. Experimental results demonstrate that this method achieves differential privacy with a privacy budget of = 1.0. The F1 scores of the synthesised data on logistic regression and decision tree tasks remain at 0.87 and 0.83, respectively. The mean absolute error (MAE) for two-dimensional association rules is 0.053 lower than that of the next-best diversity-promoting generative adversarial network (DP-GAN). This research provides an effective technical approach for high-utility privacy protection of personal information in big data environments.
  Keywords: big data; personal information anonymisation; local differential privacy; LDP; data synthesis model; privacy-utility trade-off; mean absolute error; MAE.
  DOI: 10.1504/IJIIDS.2026.10078617
   
  
• Autonomous navigation of robots in dynamic environments using reinforcement learning algorithms  
  by Zhihong Feng 
  Abstract: Due to insufficient dynamic representation of the environment and limited policy generalisation, existing methods have poor navigation performance in complex dynamic scenes. Therefore, this paper proposes a navigation method that combines hierarchical attention mechanism with meta reinforcement learning (HAMRL). This method first constructs a dual branch spatiotemporal attention network, and then utilises cross modal attention modules to fuse multi-source perceptual information, achieving robust encoding of environmental states. The experimental results show that the HAMRL method achieves a navigation success rate of 97.5% in the custom dynamic navigation benchmark. The average collision frequency is only 0.15, which is a decrease of 84.2% compared to the baseline dynamic window method (DWA) models 0.95. This high success rate is achieved while maintaining good exercise comfort, with a trajectory smoothness index of 0.018 rad/step, only 0.003 higher than the optimal DWA of 0.015.
  Keywords: reinforcement learning-based navigation; dynamic environment modelling; hierarchical attention mechanism; meta-reinforcement learning; autonomous mobile robots.
  DOI: 10.1504/IJIIDS.2026.10079049
   
  
• Trajectory tracking control system of tunnel inspection robot based on backstepping sliding mode algorithm  
  by Lingpeng Lin 
  Abstract: This paper addresses two core issues of tunnel inspection robots in crack detection tasks: the trajectory tracking accuracy is insufficient to meet the precise requirements of crack positioning and measurement, and the weak anti-interference ability in complex environments leads to a decrease in control performance. The experiment focuses on the Qinling water conveyance tunnel, and the collection period is from June to September 2023. The verification is carried out from the aspects of trajectory tracking error, anti-interference ability, control stability, and real-time performance. The results show that the RMSE of the backstepping sliding mode algorithm reaches 0.12 m; the MSE is only 0.01 m2; the MAE is only 0.08 m. Compared with the traditional PID control algorithm, the RMSE reduces the error by 0.23 m, and the response speed curve converges the fastest.
  Keywords: Qinling water conveyance tunnel; backstepping method; backstepping sliding mode algorithm; BSM; trajectory tracking control system; trajectory tracking precision.
  DOI: 10.1504/IJIIDS.2026.10079281
   
  
• User feedback classification: a case study of electric manufacturing Vietnam  
  by Kha Nguyen, Khoa Trinh, Phuoc Tran 
  Abstract: For large manufacturing enterprises, the amount of customer feedback on product quality is very large and diverse. Manually processing these feedbacks not only consumes a lot of time and resources but is also prone to errors that affect the business efficiency and reputation of the enterprise. In this paper, we focus on studying modern classification models such as T5, BERT, RoBERTa for the classification problem, specifically classifying customer feedback on the products quality. Dataset for training the classification model is collected semi-automatically from customer feedback. Experimental results show that the BERT model gives the best results among the experimental models. Based on the experimental results, we select the best model to build a customer feedback classification system for the products they use.
  Keywords: feedback classification; BERT; RoBERTa; T5; Vietnam.
  DOI: 10.1504/IJIIDS.2026.10079353
   
  
• An algorithm for mining high-occupancy patterns on weighted databases  
  by Duy Nguyen, Hang Le, Ham Nguyen 
  Abstract: High-occupancy itemset (HOI) mining is an emerging research direction in data mining that has garnered significant attention. In contrast to frequent patterns, which are measured by their occurrence frequency, HOIs are defined as itemsets that occupy a significant proportion of the transaction lengths in which they appear. Although less numerous than frequent patterns, HOIs often capture more meaningful characteristics. They are particularly useful for tasks such as data analysis and visualisation in intelligent systems. However, a key limitation of HOIs is that they only consider the presence of items, failing to account for the differences in importance or weight among them. To address this limitation, this paper introduces the concept of high-weighted occupancy patterns (HWOPs) and an efficient mining algorithm, HWOP-ROL, which leverages a novel upper bound (UBWO) to effectively prune the search space. Experimental results on various benchmark datasets demonstrate the superior efficiency of the proposed approach when compared to a baseline algorithm.
  Keywords: high-occupancy pattern; high-weighted occupancy pattern; HWOP; UBWO upper bound; HWOPROL algorithm.
  DOI: 10.1504/IJIIDS.2026.10079703
   
  
• Performance evaluation of extended long short-term memory models for financial forecasting  
  by The-Bang Nguyen, Hung C. Nguyen, Hiep Huynh, Quang-Thinh Bui 
  Abstract: Financial time-series forecasting remains challenging due to strong non-stationarity and the risk of information leakage in model evaluation. This study evaluates the extended long short-term memory (xLSTM) architecture for leakage-aware forecasting. The model integrates stabiliser memory and matrix memory to capture both short-term fluctuations and long-range dependencies while maintaining computational efficiency. A strictly time-ordered, leakage-safe evaluation protocol is adopted to ensure fair assessment. Experiments on Bitcoin and VNINDEX datasets (20172024) compare xLSTM with RNN, LSTM, GRU, and transformer models. Results indicate that xLSTM demonstrates competitive and consistent performance, achieving R2 = 0.98879 on Bitcoin and R2 = 0.95302 on VNINDEX. These findings suggest that xLSTM is a robust baseline for leakage-aware one-step-ahead forecasting under volatile market conditions.
  Keywords: deep learning; extended long short-term memory; xLSTM; conventional models; time-series; digital currency.
  DOI: 10.1504/IJIIDS.2026.10079704
   
  
• RWS-ANN: soil fertility management and crop recommendation system using machine learning in smart agricultural systems  
  by Shubhangi Vijay Gaikar, M.S. Zambare, A.D. Shaligram 
  Abstract: In smart agricultural applications, soil fertility management and crop recommendation remain the fundamental procedures, contributing to plant growth and recommending specific land for better crop yield. However, several conventional methods employed in monitoring soil fertility and crop recommendation faced drawbacks, such as inconsistent data, missing values, and reduced accuracy. As a result, the research proposes the Rider Wolf search algorithm based artificial neural network (RWS-ANN) model for accurately assessing the soil fertility and providing optimised crop recommendations using IoT technology. The proposed RWS-ANN provides precise crop recommendations on suitable lands by analysing the soil properties and the characteristics of crops. Specifically, the RWS algorithm is applied for the hyperparameter selection of the proposed RWS-ANN model, contributing to reducing the computational complexity and improving the overall accuracy. When compared with traditional methods, the proposed model achieves a mean absolute error of 1.5271, a mean squared error of 2.88, and a root mean squared error of 1.69, thus resulting in larger robustness and balanced outcomes of the research model.
  Keywords: IoT networking; soil fertility management; crop recommendation; machine learning; smart agricultural applications.
  DOI: 10.1504/IJIIDS.2026.10080057
   
  
• A depression detection framework based on sentiment analysis using random multimodal-based hierarchical deep learning  
  by Amol Govind Patil, Nitin L. Gavankar 
  Abstract: This paper proposes a convolutional neural network with transfer learning (CNN-TL) for detecting depression using review data. At first, the review data are allowed for bidirectional encoder representations from transformers (BERT) tokenisation to obtain tokens, and fed to the aspect term extraction (ATE) phase to generate aspect terms, followed by crucial features extraction. From extracted features, the classification of sentiment is executed with random multimodal hierarchical deep learning (RMHDL). Finally, classified sentiment data output, input review data and extracted features are subjected to depression discovery using a CNN-TL. Here, the CNN is used with hyperparameters considering pre-trained models such as AlexNet and VGG-16. The RMHDL performed very well, stating an elevated accuracy of 92.2%, true positive rate (TPR) of 90.7%, TNR of 92.6%, precision of 91.0% and lower mean squared error (MSE) of 10.6%.
  Keywords: depression detection; review data; random multimodal deep learning; hierarchical deep learning for text; sentiment classification.
  DOI: 10.1504/IJIIDS.2026.10080429
   
  
• Adaptive knowledge-driven query generation for zero-shot document retrieval  
  by Weiping Zou, Chenjun Sun, Chao Zheng, Bin Huang 
  Abstract: Document retrieval models often rely on large amounts of labelled data, which are time-consuming and costly to collect and annotate manually. To address this, recent methods generate pseudo queries directly from documents to train retrieval models without manual supervision. However, most existing approaches depend solely on internal document content, overlooking external knowledge that could boost performance, especially in zero-shot settings. In this paper, we propose the adaptive knowledge-driven query expansion with implicit feedback (ADDEND), a method that samples representative keywords, expands them using a knowledge graph, and refines pseudo queries to reduce redundancy and semantic drift. Evaluated on 12 publicly available benchmarks, ADDEND consistently outperforms existing models, with ablation studies confirming the effectiveness of the keyword sampling and expansion modules.
  Keywords: document retrieval; zero-shot learning; knowledge graph; internal documents; external documents.
  DOI: 10.1504/IJIIDS.2026.10080631
   
  
• Time series prediction for complex networks based on graph convolutional neural networks  
  by Xiaoli Wang 
  Abstract: By using a time-varying weighted spectral convolutional network with adaptive temporal attention fusion (TWS-GCN-ATAF) model, the challenge of integrating the topology and temporal dynamics of complex networks into spatial and temporal entities can be solved. The model also incorporates an adaptive temporal attention mechanism to effectively integrate temporal dependencies at multiple scales. The results indicate that the model provides an average mean absolute error (MAE) of 0.032 and a root mean square error (RMSE) of 0.054 for the METR-LA transportation network. The synthetic dynamic network with sudden structural changes shows a consistent average MAE of 0.039. In contrast, the Reddit social network prediction task shows an improved three-step RMSE of 0.041 and a higher trend correlation coefficient of 0.89, validating the model's prediction accuracy and stability in dynamic topological environments.
  Keywords: complex networks; spatiotemporal prediction; graph convolutional neural network; GCNN; time-varying topology; adaptive attention.
  DOI: 10.1504/IJIIDS.2026.10080743