Scholarlink

<p>Academic procrastination plays a substantial role in determining the achievements of students. This study develops a novel framework that would allow data-driven academic decisions. This study proposes a reliability-driven hyperparameter optimization framework that defines a composite optimization problem that incorporates discriminative capacity, predictive reliability, and falsified negative risks into a single framework. Another major aspect of this study is defining a prediction target labelled as engineered academic risk; this prediction target was created using structured behavioural and academic survey-based indicators. To avoid a circular and a leakage-resilient issue, preprocessing and feature selection were applied to ensure that input features were excluded from the constructed label. The goal combines ROC-AUC, a bootstrap-based stability measure of variance in the resampled data sets, and a risk-sensitive penalty that seeks to reduce the number of false alarms in identifying academically vulnerable students. Optimization of the hyperparameters utilized Optuna in stratified five-fold stratified cross-validation in order to ensure statistical robustness and reproducibility in the experiment. The experimental outcomes verify that all classifiers have a high predictive potential, with the Random Forest model showing the highest composite score of 0.9852 even as its ROC AUC stands at 0.9766, its accuracy at 92.15%, and its weighted F1-score at 0.9321, while logistic regression had a perfect recall, indicating sensitivity concerns. Statistical Significance Tests verified that the differences in the classifiers cannot be considered substantial, indicating that the reliability-focused optimization strategy was effective as opposed to a single classifier approach.</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

ACADPro: XAI-student procrastination classification in academia using optuna optimized machine learning models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月9日

查看全文
  • V. Jalaja Jayalakshmi,
  • M. Punithavalli

摘要

Academic procrastination plays a substantial role in determining the achievements of students. This study develops a novel framework that would allow data-driven academic decisions. This study proposes a reliability-driven hyperparameter optimization framework that defines a composite optimization problem that incorporates discriminative capacity, predictive reliability, and falsified negative risks into a single framework. Another major aspect of this study is defining a prediction target labelled as engineered academic risk; this prediction target was created using structured behavioural and academic survey-based indicators. To avoid a circular and a leakage-resilient issue, preprocessing and feature selection were applied to ensure that input features were excluded from the constructed label. The goal combines ROC-AUC, a bootstrap-based stability measure of variance in the resampled data sets, and a risk-sensitive penalty that seeks to reduce the number of false alarms in identifying academically vulnerable students. Optimization of the hyperparameters utilized Optuna in stratified five-fold stratified cross-validation in order to ensure statistical robustness and reproducibility in the experiment. The experimental outcomes verify that all classifiers have a high predictive potential, with the Random Forest model showing the highest composite score of 0.9852 even as its ROC AUC stands at 0.9766, its accuracy at 92.15%, and its weighted F1-score at 0.9321, while logistic regression had a perfect recall, indicating sensitivity concerns. Statistical Significance Tests verified that the differences in the classifiers cannot be considered substantial, indicating that the reliability-focused optimization strategy was effective as opposed to a single classifier approach.