Scholarlink

<p>This paper introduces an integrated Multilayer Perceptron-Decision Tree (MLP-DT) framework that improves on predictive accuracy, interpretability, and fairness of predicting student performance. The proposed architecture combines the representational ability of deep learning with the transparency of symbolic learning. Within this framework, the Multilayer Perceptron (MLP) module obtains deep, non-linear representations of features from various educational datasets that contain demographic, behavioural, and academic features, which allows the Decision Tree (DT) layer to make interpretable, rule-based decisions from these representations. This accomplishes the previously stated goal of making accurate, but opaque neural models transparent and interpretable while maintaining the limitations of conventional algorithms. The experimental results on multiple student datasets demonstrate that the hybrid model achieves better performance in terms of accuracy, F1-score, stability, and bias mitigation compared to traditional machine learning and stand-alone deep learning models. Additionally, the interpretable rules obtained from the DT layer provide actionable advice for educators facilitating early intervention, personalised learning, and fair academic help. In conclusion, the proposed framework establishes a predictive, transparent, and high-performance model for data-driven decision-making in education.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel hybrid MLP–decision tree architecture for student performance prediction: a new paradigm in educational data mining

  • 期刊论文

  • 发表日期：

    2025年12月20日

查看全文
  • Nawnit Kumar,
  • Kapil Madan

摘要

This paper introduces an integrated Multilayer Perceptron-Decision Tree (MLP-DT) framework that improves on predictive accuracy, interpretability, and fairness of predicting student performance. The proposed architecture combines the representational ability of deep learning with the transparency of symbolic learning. Within this framework, the Multilayer Perceptron (MLP) module obtains deep, non-linear representations of features from various educational datasets that contain demographic, behavioural, and academic features, which allows the Decision Tree (DT) layer to make interpretable, rule-based decisions from these representations. This accomplishes the previously stated goal of making accurate, but opaque neural models transparent and interpretable while maintaining the limitations of conventional algorithms. The experimental results on multiple student datasets demonstrate that the hybrid model achieves better performance in terms of accuracy, F1-score, stability, and bias mitigation compared to traditional machine learning and stand-alone deep learning models. Additionally, the interpretable rules obtained from the DT layer provide actionable advice for educators facilitating early intervention, personalised learning, and fair academic help. In conclusion, the proposed framework establishes a predictive, transparent, and high-performance model for data-driven decision-making in education.