Educational Data Mining (EDM) has achieved substantial gains in predictive performance, yet many existing approaches remain centered on single-objective optimization, most often accuracy. This does not adequately reflect the multi-dimensional nature of real-world educational decision-making, which requires balancing interpretability, fairness, robustness, efficiency, and timeliness. This perspective advocates a shift toward multi-objective, interpretable, and trustworthy EDM frameworks. We highlight the role of multi-objective optimization in modeling trade-offs through Pareto-optimal solutions and address the challenge of actionable decision-making through bargaining-based mechanisms, such... More >

Personalized learning has become a popular term in education to address learner diversity and enhance student performance. However, its implementation remains a practical challenge in many developing countries due to the lack of fine-grained learning data and platforms. This study proposes a data-driven machine learning approach for personalized learning using routine administrative education data from Rwanda. The approach is scalable, interpretable, and aligned with existing national education information systems. Following a design science research approach, the study combines unsupervised learner profiling via clustering and supervised performance prediction via regression models. The pro... More >

The rapid digitization of education has generated massive, heterogeneous student behavioral data, creating both opportunities and methodological challenges for educational data mining. This perspective paper discusses an emerging paradigm that integrates approximate fast clustering algorithms with deep learning techniques to enable scalable, high-resolution student behavior analysis. By leveraging approximate computation, clustering efficiency can be significantly improved without sacrificing analytical fidelity, while deep learning models facilitate the extraction of high-level representations from multimodal behavioral data. The synergy between these approaches could enable robust applicat... More >

The rapid digitalization of education has led to an explosive growth of multi-source and multi-modal learning data, providing new opportunities for advancing Educational Data Mining (EDM). By learning diverse data such as learning behaviors, assessment records, and interaction logs, EDM enables deeper insights into student learning processes and supports the development of personalized and intelligent education. However, several critical challenges remain, including the heterogeneity and fragmentation of multi-source data, the difficulty of extracting meaningful knowledge through effective data fusion, and the limited interpretability of high-performance predictive models. To address these c... More >

With the development of educational technology and the accumulation of big data, student performance prediction has become a hot topic in the field of education. However, traditional manual statistical methods have limitations in dealing with complex data and are difficult to achieve high-precision prediction. To address this gap, this study proposes a clustering-based feature generation framework to enhance prediction performance. Firstly, the multilayer perceptron (MLP) model is employed to evaluate the effectiveness of the clustering algorithms (K-Means, DBSCAN, and hierarchical clustering) for feature generation. Then, the best clustering algorithm (K-Means) is applied to generate featur... More >

Student performance prediction is a core task in educational data mining, as it enables early intervention, personalized learning support, and data-driven decision-making. Although existing machine learning models have shown promising results in this domain, challenges persist due to hard-to-classify samples—particularly students exhibiting borderline performance—and the discrete nature of hard labels, which together limit predictive effectiveness. To overcome these limitations, this paper proposes a KFWAdaBoost-based soft label learning framework that systematically enhances baseline model performance through a two-stage synergistic mechanism. In the first stage, K-means++ clustering is... More >

Predicting student dropout and academic success is important for higher education institutions for enhancing retention and deliver timely interventions. However, educational datasets often exhibit severe class imbalance, particularly when multiple academic outcomes (i.e., dropout, enrolled, and graduate) are considered simultaneously. Thus this study examines the effectiveness of three widely used over-sampling techniques (i.e., RandomOverSampler, synthetic minority oversampling technique, and adaptive synthetic sampling) for mitigating class imbalance and enhancing prediction performance. These sampling strategies are evaluated in combination with several machine learning classifiers to ass... More >

In educational data mining, accurate prediction of student performance is important for supporting timely intervention for at-risk students. However, educational datasets often include irrelevant or redundant features that could reduce the performance of prediction models. To tackle this issue, this study proposes a gradient boosting-based feature selection framework that can automatically identify and obtain the most important features for student performance prediction. The proposed framework leverages the gradient boosting model to calculate feature importance and refine the feature subset, aiming to achieve comparable or superior prediction performance using fewer but important input fea... More >