International Journal of Business Intelligence and Data Mining (12 papers in press)

Special Issue on: OA Digitalisation Information Systems and Artificial Intelligence in Business Processing

• Open Access

  Study on multimodal ideological and political teaching material push on MOOC online learning platform ( Free Full-text Access )

  by Yiming Qu, Yanhua Wang 
  Abstract: The expected goal is to address the issues of low coverage, high latency, and high average absolute error in traditional push methods for ideological and political teaching resources. This study focuses on the multimodal ideological and political teaching material push method on MOOC online learning platform. Firstly, the k-means clustering method is used to label the learner data and construct platform user profiles; Secondly, by combining significant data block detection methods, significant learning features are extracted. Using polynomial naive Bayes model, classify according to the modality of ideological and political teaching resources; Finally, based on collaborative filtering algorithms, different modalities of ideological and political teaching resources are pushed to learners with different learning needs. Through experiments, it has been proven that the coverage rate of our method can reach over 90%, with a push delay of only 105ms and an average absolute error of only 0.13.
  Keywords: MOOC online learning platform; multimodal; ideological and political teaching resources; user profiles; significant data block detection.
  DOI: 10.1504/IJBIDM.2025.10075381
   
  

• Open Access

  Prediction of carbon emissions throughout the lifecycle of zero carbon substations based on Lasso-GRNN neural network model ( Free Full-text Access )

  by Ting Zeng, Yueqing Chen, Liuhuo Wang, Mingpeng Yuan, Zhangqi Lv, Dianbin Wang 
  Abstract: In order to solve the problems of low recall rate, low prediction accuracy, and long prediction completion time of carbon emission prediction factors in traditional methods, a prediction method of carbon emissions throughout the lifecycle of zero carbon substations based on Lasso-GRNN neural network model is proposed. Lasso-GRNN neural network model is constructed by using the key indicators for predicting the carbon emissions of a zero carbon substation throughout its entire lifecycle as input variables and the carbon emission values as output variables. The model uses Lasso to screen key indicators and inputs them into the GRNN neural network to obtain accurate prediction results. Experimental results show that the proposed method has a maximum recall rate of 98.12% for the influencing factors of carbon emissions throughout the entire life cycle of zero carbon substations, a maximum prediction accuracy of 98.51%, and a minimum prediction completion time of 0.68s.
  Keywords: Lasso-GRNN neural network model; zero carbon substation; full lifecycle; carbon emissions forecast; indicators.
  DOI: 10.1504/IJBIDM.2026.10076223
   
  

Special Issue on: OA Digitalisation, Information Systems and Artificial Intelligence in Business Processing

• Open Access

  Research on safety risk perception of electochemical energy storage power station under the background of environmental sustainable development ( Free Full-text Access )

  by DongLiang Deng, LiXin Yin 
  Abstract: The current risks faced by electrochemical energy storage power plants are diversified and complex, resulting in high false alarm rates, false negative rate, and long time consumption of traditional methods. Therefore, a safety risk perception method of electochemical energy storage power station under the background of environmental sustainable development has been proposed. Cluster analysis of multi-source risk indicator data using GMM to identify potential risk patterns; It calculates the weights of various indicators based on cloud models, screens key risk factors, and constructs a multi-level risk indicator system; by integrating real-time monitoring data with prior knowledge through Bayesian inference, dynamic risk probability updates and perception can be achieved. The experimental results show that the minimum false alarm rate of the proposed method is 2.86%, the minimum false negative rate is 2.78%, and the risk perception time varies between 0.3 s and 0.7 s, indicating high engineering application value.
  Keywords: background of environmental sustainable development; cloud model; electochemical energy storage power station; safety risk perception; Bayesian inference.
  DOI: 10.1504/IJBIDM.2025.10074955
   
  

• Open Access

  A comprehensive management method of audit databased on knowledge graph ( Free Full-text Access )

  by Xuena Lin 
  Abstract: A comprehensive management method of audit data based on knowledge graph is proposed to solve the problems of low F1 value, long data update delay time, and low data coverage in traditional methods. Firstly, using web crawling technology to automate the collection of audit data. Secondly, based on the preprocessed data, a BiLSTM CRF joint model is used to achieve audit entity recognition, and a graph convolutional network (GCN) is used to complete the relationship extraction task, thereby constructing an audit knowledge graph. Finally, an incremental learning mechanism is introduced to dynamically update the knowledge graph, and comprehensive management of audit data is achieved based on the updated knowledge graph. The experimental results show that the F1 value of the proposed method is between 0.81 and 0.89, the data update delay time is stable at 100-110ms, and the data coverage reaches over 90% after 10 iterations and remains stable
  Keywords: knowledge graph; KG; audit data; entity recognition; relationship extraction; dynamically update.
  DOI: 10.1504/IJBIDM.2025.10075009
   
  

Special Issue on: Knowledge Discovery from Big Data to Spur Social Development

• Leveraging traditional business culture for business intelligence: a scalable parameter server architecture with distributed machine learning  
  by Chengcai Xing 
  Abstract: Yanan, a significant historical and cultural hub in China, is being revitalised and utilised to drive development in various spheres. The citys traditional commercial and cultural resources are being harnessed to contribute to its political, cultural, educational, and economic growth. Yanan models other historically significant regions, demonstrating how heritage can be leveraged for contemporary development. Advanced machine learning approaches are used to overcome scalability and robustness issues in large-scale data-driven systems. The parameter server architecture decentralises the training process of machine learning models, enabling efficient handling of vast datasets and high computational demands. This design enhances fault tolerance and ensures seamless operation under challenging conditions. Intelligent simulations and tests validate the efficacy of these machine-learning approaches in modelling the evolution and application of traditional commercial culture. These simulations provide a dynamic and accurate representation of how cultural and business practices can adapt and thrive in modern contexts. The reliability and precision of machine learning models in capturing complex patterns and trends inherent in cultural and economic transitions are underscored through rigorous testing. This exploration highlights the innovative intersection of technology and tradition, showcasing how machine learning can play a transformative role in preserving and advancing historical and cultural assets.
  Keywords: traditional commercial and cultural resources; Yan’an historical value; distributed machine learning; parametric server architecture.
  DOI: 10.1504/IJBIDM.2025.10069536
   
  
• Deep learning based DAVS-UNet for medical image segmentation  
  by Zihui Zhu 
  Abstract: Deep learning based Convolutional Neural Networks (CNNs) and transformers are widely used in medical image processing tasks, while the State Space Sequence Model (SMM) architecture is proposed to address its limitations in improving the scaling efficiency and solving the transformed quadratic scale problem. Inspired by the Mamba architecture, this paper proposes Dual-Attention Vision Scaled-UNet (DAVS-UNet) for medical image segmentation, in which Adaptive Multi-scale Selection (AMS) is applied to the input image for better capturing details at different scales and extracting input features. Furthermore, Atrous Space Pyramid Pooling (ASPP) is introduced to expand the sensory field by collecting global contextual information after the final encoder. The experiments on a large number of publicly available datasets illustrate that DAVs-UNet shows excellent performance on the ISIC2017, ISIC2018, Synapse datasets, and outperforms existing SSM-architecture networks employed in medical image segmentation tasks. The code is available at https://github.com/zhzhuac/DAVS.
  Keywords: attention mechanism; multi-scale information; state space models; convolutional neural networks; CNNs; adaptive multi-scale selection; AMS; atrous space pyramid pooling; ASPP.
  DOI: 10.1504/IJBIDM.2026.10070330
   
  
• GA-TabNet: a novel approach for early dropout prediction in MOOCs based on genetic algorithms and TabNet  
  by Houssam Eddine Aouarib, Fatima Zohra Laallam, Salah Eddine Henouda, Mohamed Fouzi Djouhri 
  Abstract: Massive open online courses (MOOCs) represent one of the most effective educational methodologies due to their cost-effectiveness, flexibility, ubiquity, and their role in facilitating and improving education. MOOCs possess the capacity to revolutionise global education; nevertheless, the high dropout rates often undermine their effectiveness. The emergence of machine learning, deep learning techniques, and educational big data enables academics to address the student dropout problem through big data analytics. This study addresses the critical challenges of student dropout prediction by proposing GA-TabNet, an innovative model that combines a Genetic Algorithm with TabNet for early dropout prediction. The results of this study were validated using the Open University Learning Analytics Dataset. The proposed model attained an average accuracy exceeding 92%. Furthermore, it outperformed traditional predictive models, including Support Vector Machine, Long Short-Term Memory, Logistic Regression, MultiLayer Perceptron, Decision Trees, and Random Forest models, by margins ranging from 0.79% to 4.79%.
  Keywords: student dropout prediction; massive open online course; MOOC; TabNet; genetic algorithm; big data; machine learning; ML; deep learning; DL.
  DOI: 10.1504/IJBIDM.2026.10073518
   
  
• Enhancing physical education with kinect motion tracking and context personalisation  
  by Yuqiu Zhang 
  Abstract: This study proposes a novel approach for physical education (PE) that integrates kinect motion tracking, deep learning, and context personalisation. The system combines real-time feedback and adaptive learning paths to optimise student participation, motivation, and physical skill development. An ablation study was conducted to compare the effectiveness of the full system with three other configurations: kinect-only motion tracking, kinect with context personalisation, and kinect with deep learning. The experimental results indicate that the full system, which combines all three components, significantly outperforms the other configurations in terms of motivation, physical performance improvement, and engagement. Specifically, the full system achieved the highest improvement in skill development (90%), engagement (98%), and motivation, suggesting that the combination of kinect motion tracking, context personalisation, and deep learning is most effective for enhancing PE outcomes. This research contributes to the digital transformation of physical education. It provides a new pathway to leverage technology for improving both student motivation and performance.
  Keywords: physical education; kinect; deep learning; context personalisation; teaching methods.
  DOI: 10.1504/IJBIDM.2026.10074262
   
  
• Integrating data mining and transformer models for socially beneficial bicycle scheduling in urban bicycle-sharing  
  by Wenxin Ma, Jiabin Liang, Dazhou Li 
  Abstract: The rapid growth of bicycle-sharing systems in urban settings necessitates the implementation of effective scheduling strategies to optimise resource allocation and tackle challenges such as uneven distribution of bicycles and operational inefficiencies. This study introduces an innovative integration of data mining techniques and transformer models aimed at enhancing bicycle scheduling. By leveraging the MARO resource scheduling platform, we simulate bicycle mobility and scrutinise demand patterns through electronic fence clustering. We evaluate two distinct scheduling strategies: a dynamic programming-based approach and a transformer-based methodology. Experimental findings reveal that the proposed transformer model markedly decreases average time overhead and path distances in comparison to conventional methods, thereby fostering more efficient and socially advantageous bicycle-sharing systems. This research significantly contributes to the optimisation of urban transportation and the advancement of sustainable mobility solutions.
  Keywords: bicycle scheduling; data mining; transformer model; urban bicycle-sharing; resource optimisation.
  DOI: 10.1504/IJBIDM.2026.10074963
   
  
• Enhanced random forest-based teaching practice and analysis for learning behaviour prediction  
  by Hongzhi Wei 
  Abstract: Educational data mining has emerged as a pivotal discipline for enhancing pedagogical practices through data-driven insights. While significant progress has been achieved in analysing learning behaviours, accurate prediction of teaching practices and student dynamics remains challenged by data complexity, behavioural variability, and methodological limitations of conventional approaches. This paper presents an enhanced random forest-based predictive method that systematically addresses these challenges through three key innovations. First, advanced feature selection mechanisms for handling high-dimensional educational data; Second, optimised ensemble learning architecture improving prediction reliability; Third, multi-source data integration capabilities enabling comprehensive behavioural analysis. Moreover, from the experimental results, the proposed method demonstrates over 15% higher accuracy and over 20% improvement compared to the state-of-the-art methods in terms of both the performance and cross-validation evaluations.
  Keywords: data fusion; teaching practice; random forest; behaviour prediction.
  DOI: 10.1504/IJBIDM.2026.10076096
   
  
• Attention and deep feature-based intelligent approach for abnormal network traffic detection  
  by Guihua Wu 
  Abstract: Network traffic anomaly detection faces critical challenges in feature extraction robustness, and computational efficiency due to increasing data dimensionality and environmental noise. Existing deep learning approaches offer partial improvements but suffer from noise sensitivity, structural information neglect, and unnecessary computational overhead. This paper presents an intelligent approach integrating attention mechanisms with deep feature learning, combining multi-scale attention dynamics and optimised gradient boosting to address network anomaly detection challenges. The core contributions encompass a hybrid solution that achieves noise-resilient feature extraction through self-attention weighting while preserving structural traffic patterns, coupled with an enhanced and optimised gradient boosted decision tree classifier employing logarithmic loss optimisation and early stopping mechanisms for effective high-dimensional sparse data processing. Experimental results demonstrate that the proposed approach outperforms the state-of-the-art baselines, including superior detection accuracy 98.2% and around 34.7% detection time reduction.
  Keywords: abnormal traffic detection; self-attention weighting; deep feature; intelligent approach; network traffic.
  DOI: 10.1504/IJBIDM.2026.10076097
   
  
• A deep learning approach to classifying ECG signals under motion artefacts  
  by Long Xu 
  Abstract: Wearable electrocardiogram (ECG) monitoring is essential for health assessment during physical activity; however, motion artifacts remain a major challenge, severely compromising signal quality. Traditional filtering techniques often lack adaptability to dynamic noise variations, and deep learning models typically overlook motion-induced signal degradation. To overcome these limitations, we propose an acceleration-assisted framework for ECG denoising and activity classification. Specifically, motion intensity is calculated from chest acceleration data, guiding the adaptive removal of intrinsic mode functions (IMFs) via empirical mode decomposition (EMD). The denoised ECG signals are subsequently classified using a one-dimensional convolutional neural network (1D-CNN). Experiments conducted on the publicly available MHEALTH dataset demonstrate the effectiveness of the proposed method, yielding classification metrics of 0.485 (accuracy), 0.492 (precision), 0.486 (recall), and 0.476 (F1- score). The framework effectively mitigates motion artifacts while preserving essential ECG waveform features, ensuring more robust performance across various physical activities.
  Keywords: electrocardiograph signal; ECG; motion artefact; empirical mode decomposition; EMD; acceleration signal; 1D convolutional neural network; activity recognition.
  DOI: 10.1504/IJBIDM.2026.10076364