Scholarlink

<p>The study integrated machine learning (ML) framework for predicting students’ success in academics via 2-phased experiments with emotional intelligence and personality traits. In the initial experiment, unsupervised learning (K-Means clustering) was utilized to unearth hidden success levels from unlabeled data. This precedes training of standalone ML classifiers [Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost)] on clustered-derived labels. Phase 2 experiment; ensemble model was developed by integrating the standalone ML models via stacking and voting using Logistic Regression as the meter-learner. Models’ evaluation was done with Accuracy, F1-score, Recall, AUC-ROC, and Precision. The proposed ensemble model recorded optimal results for all the metrices, outperforming standalone models. Comparative analysis with previous studies validated the superior generalizability of the proposed model. Decision Tree was used to enhance interpretability, unveiling Conscientiousness, Stress Management, and Trait Empathy as dominant predictors. The study’s findings emphasized the relevance of integrating affective and personality factors in educational analytics as well as the need for interpretable, ensemble-based predictive tools for early risk detection to enhance decision-making in pursuance to students’ academic success rate.</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

Intelligent profiling beyond grades using a stacking ensemble framework for student success prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月20日

查看全文
  • Samuel Odoom,
  • Eric Opoku Osei,
  • Enock Quansah Effah,
  • Amiru Bakariwie,
  • Ahmed Adams Rufai

摘要

The study integrated machine learning (ML) framework for predicting students’ success in academics via 2-phased experiments with emotional intelligence and personality traits. In the initial experiment, unsupervised learning (K-Means clustering) was utilized to unearth hidden success levels from unlabeled data. This precedes training of standalone ML classifiers [Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost)] on clustered-derived labels. Phase 2 experiment; ensemble model was developed by integrating the standalone ML models via stacking and voting using Logistic Regression as the meter-learner. Models’ evaluation was done with Accuracy, F1-score, Recall, AUC-ROC, and Precision. The proposed ensemble model recorded optimal results for all the metrices, outperforming standalone models. Comparative analysis with previous studies validated the superior generalizability of the proposed model. Decision Tree was used to enhance interpretability, unveiling Conscientiousness, Stress Management, and Trait Empathy as dominant predictors. The study’s findings emphasized the relevance of integrating affective and personality factors in educational analytics as well as the need for interpretable, ensemble-based predictive tools for early risk detection to enhance decision-making in pursuance to students’ academic success rate.