International Journal of Business Intelligence and Data Mining (23 papers in press)

Regular Issues

• Improving brain MRI segmentation of multiple sclerosis using an advanced CNN approach  
  by V. Biksham , Sampath Korra, B. Pradeep Kumar , Salar Mohammad 
  Abstract: Multiple sclerosis (MS) can be detected early by looking for lesions in brain magnetic resonance imaging (MRI). Recently, unsupervised anomaly detection algorithms based on autoencoders were presented for the automatic identification of MS lesions. However, because these autoencoder-based approaches were created exclusively for 2D MRI pictures (e.g., 2D cross-sectional slices), they do not take use of the complete 3D information of MRI. In this research work, a novel 3D autoencoder-based methodological solution for detecting MS lesion volume in MRI is offered. We begin by defining a 3D convolutional neural network (CNN) for complete MRI volumes and then construct each encoder and decoder layer of the 3D autoencoder using 3D CNN. For optimal data reconstruction, we additionally include a skip link between the encoder and decoder layers. In the experimental results, we compare the 3D autoencoder-based method to the 2D autoencoder models using training datasets from the Human Connectome Project (HCP) and testing datasets from the Longitudinal MS Lesion Segmentation Challenge, and show that the proposed method outperforms the 2D autoencoder models by up to 20% in MS lesion prediction.
  Keywords: multiple sclerosis; brain MRI; image segmentation; CNN; chronic disease; healthcare.
  DOI: 10.1504/IJBIDM.2025.10068668
   
  
• Analysing the effectiveness of financial news sentiments on stock price prediction of twelve Indian sectoral stock indices using a hybrid LSTM-GRU model  
  by Meera George, R. Murugesan 
  Abstract: Despite the growing interest in combining news sentiments with historical data to improve stock price prediction, a considerable gap exists in predicting the sectoral stock indices using this methodology. This study addresses this gap by predicting the closing price of 12 Indian sectoral stock indices through a hybrid deep-learning architecture. It employs a hybrid TFIDF-Doc2Vec feature extraction technique and an SVM classifier to extract the financial news sentiments. These sentiments are utilised to create a sentiment Index, combined with historical stock data to predict each sectoral stock index using a hybrid LSTM-GRU model. The study evaluates the effectiveness of financial news sentiments in sectoral stock prediction by comparing models with and without sentiments. Results demonstrate a notable influence of sentiments on the stock price prediction of ten sectoral stock indices with a pronounced impact on the NSEBANK index. This study offers valuable insights for investors in formulating sector-specific trading strategies. It also aids policymakers in market regulation and helps financial analysts improve forecasting models by incorporating financial news sentiments. Future research could explore the integration of multi-source investor sentiments with advanced deep-learning models for even more accurate stock price predictions across diverse sectors.
  Keywords: stock price prediction; sectoral stock indices; financial news sentiments; FNSs; hybrid TFIDF-Doc2Vec; hybrid LSTM-GRU.
  DOI: 10.1504/IJBIDM.2025.10070322
   
  
• Data mining of abnormal behaviour in English learning based on dynamic grid generation algorithm  
  by Le Yang  
  Abstract: To address the limitations of traditional data mining methods for detecting abnormal behaviour in English learning - such as low accuracy, low recall rates, and long processing times - this paper proposes a joint data mining method based on a dynamic grid generation algorithm. The proposed approach begins by collecting instances of abnormal behaviour in English learning to construct a behavioural dataset. Next, the RANSAC algorithm eliminates erroneous feature matching points, while the dynamic grid generation algorithm performs data feature matching. Finally, an abnormal behaviour data mining function is constructed to identify anomalies in English learning behaviours. Experimental results demonstrate the methods effectiveness, achieving a recall rate of 96.5%, an accuracy of 99.15%, and a processing time of just six seconds, confirming its efficiency in mining abnormal behaviour data in English learning.
  Keywords: feature matching; dynamic grid generation algorithm; data mining; Ransac algorithm.
  DOI: 10.1504/IJBIDM.2025.10072627
   
  

Special Issue on: Data Analysis and Mining in Business Domains New Techniques and Applications

• A method for mining comment text data on e-commerce platforms for enterprise digital transformation  
  by Yang Wang 
  Abstract: In order to solve the problems of poor data mining performance, high RMSE value, and low stability variance in traditional e-commerce platform comment text data mining methods, a new method for mining comment text data on e-commerce platforms for enterprise digital transformation is proposed. Firstly, LDA is used for dimensionality reduction of comment text data; Secondly, the information gain method is applied to screen the key features of the comment text data based on the dimensionality reduction results; Finally, based on the selected key features, the K-means algorithm is used to cluster and mine the comment text data. The experimental results show that this method can effectively identify and classify different types of comment data, and obtain more accurate mining results. After multiple iterations, its RMSE index remained stable at around 0.2, and the highest stability variance reached 98.3, indicating that its data mining results were more accurate and stable.
  Keywords: enterprise digital transformation; e-commerce platform; text data mining; LDA; key features; k-means algorithm.
  DOI: 10.1504/IJBIDM.2025.10069901
   
  
• Book classification recommendation method for university libraries based on collaborative filtering algorithm  
  by Yina Liu 
  Abstract: The recommendation of book classification in university libraries is of great significance for improving information retrieval efficiency and optimising book resource management. In order to solve the problems of low accuracy, long time, and low user satisfaction in traditional book classification recommendation methods for university libraries, a book classification recommendation method for university libraries based on collaborative filtering algorithm is proposed. Firstly, the fuzzy C-means clustering algorithm is used to cluster the data of university library platforms, completing the collection of library platform data. Secondly, determine the user characteristics of university libraries based on the collected data. Finally, the collaborative filtering algorithm calculates the predicted scores of university library books based on user characteristics, and implements book classification recommendations for university libraries. Experimental results show that the maximum classification recommendation accuracy of the proposed method is 97.6%, the average recommendation time is 0.52 s, and the average user satisfaction is 94.71.
  Keywords: collaborative filtering algorithm; university library; book classification recommendation; fuzzy C-means clustering algorithm; user characteristics.
  DOI: 10.1504/IJBIDM.2025.10069902
   
  
• Rapid detection of abnormal sales data on e-commerce platforms under the digital transformation of enterprises  
  by Jing Wang, Hongmei Zhao 
  Abstract: In order to quickly and accurately detect outliers in sales data, a new e-commerce platform sales data abnormal rapid detection method is proposed under the digital transformation of enterprises. Firstly, analyse the impact of enterprise digital transformation on sales data of e-commerce platforms. Secondly, principal component analysis is used to extract key information through dimensionality reduction techniques, reducing computational complexity. Singular value decomposition is utilised to process data and effectively identify the main factors affecting sales. Again, by calculating the dispersion of sales data, quantitatively evaluate the fluctuation of sales data. Finally, optimize the grid partitioning and KNN algorithm parameters, and use the fast density peak algorithm to achieve efficient and real-time abnormal detection in e-commerce platform sales data. The experimental results show that the data abnormal detection accuracy of our method consistently remains above 91%, and the longest detection time does not exceed 10 seconds.
  Keywords: digital transformation of enterprises; e-commerce platform; sales data; rapid detection of anomalies.
  DOI: 10.1504/IJBIDM.2025.10069903
   
  
• Parsing and verification method of basic power grid data based on multi data source fusion  
  by Zhibin Zhou, Zhiguo Zhou, Xiongfeng Ye 
  Abstract: In order to solve the problems of low parsing accuracy, low verification accuracy, and long data parsing and verification time in traditional power grid basic data parsing and verification methods, a parsing and verification method of basic power grid data based on multi data source fusion is proposed. Using the D-S evidence theory to fuse multiple data sources in the power grid, smoothing the fusion results and inputting them into an RBF neural network to obtain the parsing results of the power grid basic data. Combining the five verification principle attributes and Bayes’ theorem, the power grid basic data verification is implemented. The experimental results show that the average data parsing accuracy of the proposed method is 96.69%, the average validation accuracy is 96.48%, and the time consumption varies between 0.23s and 0.55s, which is of great significance for improving data quality and management level.
  Keywords: multi data source fusion; basic power grid data; parsing verification; D-S evidence theory; RBF neural network; Bayes’ theorem.
  DOI: 10.1504/IJBIDM.2025.10069904
   
  
• Study on cloud resource scheduling in power multi service scenarios based on large language model technology framework  
  by Shuhong Wu 
  Abstract: Due to the complexity of various business scenarios in the power industry, it is difficult to achieve load balancing, resulting in long cloud resource scheduling and system execution times. Propose a cloud resource scheduling algorithm for power multi service scenarios based on the big language model technology framework. Using triangular fuzzy number analysis to determine the uncertainty of execution time, and using logarithmic method to unify the data scale, the optimisation objective of cloud resource scheduling is determined. Using the linear variation of sine functions in big language modelling techniques to determine scheduling order. By utilising multi head self attention and feedforward neural networks for internal transmission, a pre trained model is constructed, and combined with fine-tuning and implementation stages, cloud resource scheduling is achieved. Experiments have shown that this algorithm reduces the execution time and cost of cloud resource scheduling in multi service scenarios of electricity.
  Keywords: large language model; power multi service scenario; cloud resource scheduling; two level mode; internal transmission.
  DOI: 10.1504/IJBIDM.2025.10070230
   
  
• Personalised recommendation of English MOOC teaching resources based on multi dimensional user portrait  
  by Yingying Zhu, Jipeng Mao 
  Abstract: In order to improve the accuracy of personalised recommendation of English MOOC teaching resources, the research on personalised recommendation method of English MOOC teaching resources based on multi-dimensional user portrait was carried out. This paper first introduces the self attention mechanism, extracts the user attribute features and potential features, and integrates them to realise the construction of multi-dimensional user portraits, and then considers the basic attributes, interest attributes and social attributes to complete the calculation of user similarity. Finally, on this basis, combined with the attribute characteristics of teaching resources and user interest characteristics, it completes the personalised recommendation of English MOOC teaching resources. The experimental results show that the accuracy of the proposed method is higher than 96.3%, the recall rate is higher than 95.7%, and the F1 value is between 0.93~0.98002E.
  Keywords: user portrait; English courses; MOOC teaching resources; personalised recommendation.
  DOI: 10.1504/IJBIDM.2025.10070231
   
  
• A performance analysis of enterprise human resource management from the perspective of social relationship network analysis  
  by Li-li Ji, Yinting Zhao 
  Abstract: To enhance the efficiency of enterprise human resource management, this study proposes a performance evaluation method based on social network analysis (SNA). Firstly, relationship data among employees were collected to construct an intra-enterprise social network model from the perspective of social network analysis. Subsequently, leveraging the analysis of social network characteristics, a performance evaluation index system was established. Finally, data envelopment analysis (DEA) was employed to evaluate the performance of enterprise human resource management. Case study results demonstrate that the implementation of the proposed method led to significant improvements across multiple aspects of human resource management, confirming its effectiveness.
  Keywords: social relationship network; human resources; performance evaluation; validity coefficient; data envelopment analysis; DEA.
  DOI: 10.1504/IJBIDM.2025.10072611
   
  
• Fast classification method for online vocational education resources based on big data decision tree  
  by Ruiron Lan, Shanshan Yu 
  Abstract: To solve the problems of low efficiency and accuracy in the rapid classification of vocational education online resources, a fast classification method for vocational education online resources based on big data decision trees is proposed. Firstly, the effective information return state formula is used to filter valuable information, and wavelet transform is used to remove data noise. Secondly, the sliding window method is used to increase teaching resource data, calculate the mutual information value between features and learning objectives, and select high mutual information features to form the final feature set. Finally, using C4.5 decision trees, calculate information entropy, information gain, etc., to achieve effective classification of online vocational education resources. Experimental results have shown that the convergence speed of the classification method proposed in this paper is fast, consistently below 0.4 seconds, with an AUC area close to 1, and a classification accuracy consistently above 90%.
  Keywords: decision tree algorithm; vocational education; online resources; information entropy; information gain rate.
  DOI: 10.1504/IJBIDM.2025.10072612
   
  
• Influence of innovation and entrepreneurship choices of college students based on social network analysis  
  by Fanwei Zeng, Haikun Li 
  Abstract: This article investigates the impact of college students’ innovation and entrepreneurship choices using social network analysis. Firstly, the characteristics of social networks were identified, and then 14 hypotheses were proposed regarding social networks, entrepreneurial intentions, entrepreneurial self-efficacy, types of social networks, relationships between entrepreneurial types, and entrepreneurial environments; Finally, a survey questionnaire was designed for data collection, measure the reliability of the questionnaire and factor analysis through internal consistency coefficient, sampling suitability test, and Bartlett sphericity test, and analyse the correlation between variables. Through experiments, it has been proven that the compatibility and comprehensiveness of using the method proposed in this article to analyse the impact of college students’ innovation and entrepreneurship choices are both above 90%. The hypotheses and conclusions are consistent, and the analysis effect is good.
  Keywords: social network analysis; college students; innovation and entrepreneurship; self-efficacy; entrepreneurial intention.
  DOI: 10.1504/IJBIDM.2025.10072613
   
  
• Method for recommending the best sightseeing route for tourist attractions based on machine learning algorithms  
  by Fengjuan Tian, Weina Pei 
  Abstract: In order to improve the coverage of tourist attractions and user satisfaction, a machine learning algorithm based method for recommending the best sightseeing route for tourist attractions is proposed. Firstly, comprehensively analyse user needs through three aspects of scenic spot ratings (popularity, user travel time, and scenic spot travel time). Secondly, utilising hidden Markov models in machine learning to process sequential data of tourist travel behaviour, mining potential patterns to predict tourist attraction choices, and constructing an best sightseeing route recommendation model that fits tourist preferences. Finally, based on the recommended route values, combined with tourists' interests and time budget, the best sightseeing route is selected. The experimental results show that the method proposed in this paper consistently maintains a coverage rate of over 80% for tourist attractions along the sightseeing route, and consistently maintains a satisfaction rate of over 90% for tourists.
  Keywords: machine learning; scenic spot; recommended best sightseeing route; hidden Markov model; HMM.
  DOI: 10.1504/IJBIDM.2025.10072614
   
  
• A method for pushing remote teaching materials based on knowledge graph and user profile  
  by Linlin Hong, Meilian Jiang 
  Abstract: In order to improve the accuracy of teaching resource push and reduce push latency, a remote teaching material push method based on knowledge graph and user profile is proposed. Firstly, the graph is divided into subsets to search for similar users. Based on a multi-layer RNN entity prediction model, the similarity probability evaluation is obtained through multi-layer processing to complete the target platform user relationships. Secondly, the data is collected from multiple sources in a spatial and temporal manner, and the DVMD denoising algorithm is used to denoise it, forming a personalised user profile. Finally, a remote teaching resource accurate push model based on support vector machine was constructed by combining user profiles, achieving real-time push of personalised learning resources. The experimental results show that our method consistently maintains a high level of over 96% in accuracy testing, and is much lower in push latency compared to other methods.
  Keywords: knowledge graph; user profile; remote teaching; teaching material push.
  DOI: 10.1504/IJBIDM.2025.10072615
   
  
• Multi label innovation and entrepreneurship data classification method based on data mining  
  by Xinyue Hou 
  Abstract: To address the challenges of low efficiency and poor accuracy in classifying innovation and entrepreneurship data, this study proposes a multi-label classification method for innovation and entrepreneurship flow databased on data mining techniques. The methodology begins with data acquisition through web crawling technology, followed by comprehensive data cleaning. Subsequently, stream data features are extracted using TF-IDF in data mining techniques. A multi-label classification model is then constructed by integrating Chinese webpage classification information through the PageCNN architecture. The final stage involves consolidating classification results from multiple single-label classifiers to generate multi-label outputs. Experimental results demonstrate that the proposed method achieves a classification accuracy of 97.1% with a processing time of only 2.1 s, significantly improving both the efficiency and effectiveness of innovation and entrepreneurship flow data classification.
  Keywords: data mining; multi label classification; innovation and entrepreneurship; PageCNN model; classifier.
  DOI: 10.1504/IJBIDM.2025.10072616
   
  
• Multivariate regression analysis on the influencing factors of Internet financial products investment  
  by Suying Nian 
  Abstract: In order to improve the comprehensiveness and contribution rate of regression analysis of influencing factors of internet financial product investment, from the perspective of behavioural finance, expected return, perceived risk, personal preference, conformity psychology, perceived usefulness and perceived ease of use are determined as independent variables, and eight hypotheses are put forward; Normalise the collected data and solve the problem of data imbalance through SMOTE algorithm; Construct a logistic regression model and test its effectiveness through confusion matrix, ROC curve, and fitting. Through experiments, it has been proven that the variance contribution rate of the factors analysed by the method proposed in this article always remains above 90%, and the comprehensiveness of factor analysis always remains above 85%. This shows that the factors selected in this paper by using the Logistic model have a greater impact on Internet financial product investment, and the regression analysis is comprehensive and robust.
  Keywords: internet financial management; investment influencing factors; multiple regression; logistic regression; confusion matrix.
  DOI: 10.1504/IJBIDM.2025.10072617
   
  
• Accurate identification of health status of substation equipment based on multi-source data fusion  
  by Li Zhang, Junjie Zhang, Xinzhuo Li, Lei Zheng, Yuxiao Zhang 
  Abstract: To improve the accuracy of substation equipment fusion analysis and reduce the failure rate of substation equipment, a precise identification method for the health status of substation equipment based on multi-source data fusion is proposed. Firstly, the multi-source data of substation equipment status information are standardised, including data classification, encoding, and coordinate system transformation. Secondly, by utilising the nonlinear mapping capability of the RBF neural network and the weight allocation of the OWA operator, the effective fusion of multi-source data information for substation equipment can be achieved. Finally, by introducing correlation coefficients, constructing a Model-1 feature domain, and utilising Copula function classification combined with the ID3 algorithm decision tree, accurate identification and classification of the health status of substation equipment were achieved. The test results demonstrate that the proposed method achieves a maximum data matching accuracy of 0.95 and reduces the failure rate of substation equipment to below 2%.
  Keywords: multi-source data fusion; substation; equipment health status; accurate identification.
  DOI: 10.1504/IJBIDM.2025.10072620
   
  
• Personalised push of ideological and political network teaching resources based on user characteristics  
  by Jinzhe Zhang 
  Abstract: To address the problems of low accuracy and AUC value in traditional personalised push methods for ideological and political network teaching resources, a personalised push method of ideological and political network teaching resources based on user characteristics is proposed. By leveraging a knowledge graph to analyse user preference data, in order to extract user preference features. Using the extracted preference features and combined with the information of ideological and political teaching resources, a model for pushing ideological and political network teaching resources is constructed, which includes four modules: resource encoding layer, candidate perception layer, multi interest extraction layer, and click prediction layer, to achieve the function of resource pushing. The experimental results show that the push accuracy of the proposed method is higher than 94.9%, and the AUC value is of 0.894, indicating strong practical applicability.
  Keywords: user characteristics; political; network teaching resources; personalised push.
  DOI: 10.1504/IJBIDM.2025.10072622
   
  
• Parallel clustering method for complex attribute network big data based on transfer learning  
  by Zhicheng Jia 
  Abstract: Complex attribute network data has high-dimensional features, with a large number of redundant or weakly correlated attributes, which increases computational complexity, increases storage requirements and computation time, and affects the efficiency and accuracy of clustering. Therefore, a parallel clustering method for complex attribute network big data based on transfer learning is proposed. In the data preprocessing stage, high-dimensional load attribute network data undergoes dimensionality reduction, denoising, and standardisation to reduce data complexity and improve data quality. We introduce deep learning technology and transfer learning theory to construct a parallel clustering analysis model, enabling rapid processing and analysis of large-scale data. The experimental results show that the clustering accuracy of the proposed method is higher than 0.94, and the NML value is higher than 0.95, effectively improving clustering performance, demonstrating greater application value, and providing a new approach for network big data clustering.
  Keywords: transfer learning; complex attribute network; big data mining; parallel clustering.
  DOI: 10.1504/IJBIDM.2025.10072623
   
  
• The influencing factors of consumers purchase behaviour preference in the internet market based on joint analysis  
  by Meijuan Li 
  Abstract: This study employs conjoint analysis to examine the determinants of consumer purchase behaviour preferences in digital market environments. The research methodology involves decomposing products into key attributes and collecting corresponding consumer preference data, followed by determining full-profile preferences and calculating attribute importance weights. Subsequently, conjoint analysis is applied to estimate simulated product market shares, thereby quantifying the influence of different attributes on consumer purchase decisions. The Bradley-Terry-Ross model is then implemented to compute attribute utility values and validate the findings. The analysis reveals that logistics distribution systems and e-commerce platform features constitute the primary factors influencing purchase behaviour preferences. Through iterative recommendation processes, the proposed method achieves a peak purchase prediction accuracy of 88.1%, demonstrating its efficacy in forecasting consumer preferences with substantial precision.
  Keywords: joint analysis method; consumer; purchase preference behaviour; internet market; purchase decision.
  DOI: 10.1504/IJBIDM.2025.10072624
   
  
• Anomaly detection method for geological hazard time-series data based on multi-source data fusion  
  by Cao Lu, Yang Fang, Linchao Huang, Wenpu Wang, Liya Ji 
  Abstract: In order to improve the integrity of geological hazard time-series data and enhance the accuracy of anomaly detection, a multi-source data fusion method for anomaly detection of geological hazard time-series data is proposed. Quantify the uncertainty of multi-source data fusion of geological hazards through D-S evidence theory and obtain the fusion results. Construct a geological hazard time-series data anomaly detection model, including anomaly models and incremental learning modules, using autoencoders, graph attention networks, and gated recurrent units to capture feature correlations and temporal characteristics. The model identifies anomalies through reconstruction error, multi head attention mechanism, and gating mechanism, calculates the root mean square error as the anomaly score, and compares it with the preset value to determine anomalies. The test results show that the data integrity of our method consistently remains above 90%, and the highest anomaly detection accuracy reaches 98.45%.
  Keywords: multi-source data fusion; geological hazards; time series data; anomaly detection.
  DOI: 10.1504/IJBIDM.2025.10072626
   
  

Special Issue on: Innovative AI driven 3D Modelling for Business Intelligence Tools

• 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
   
  

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

• 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