Scholarlink

<p>This study leverages AI to predict undergraduate performance after first semester and unveil potent insights for refining admission criteria. We move beyond the siloed gaze of mere marks and entry tests, demonstrating that incorporating academic performance in specific subjects alongside demographic factors like gender, father’s profession, and exam board unlocks significantly improved early performance prediction. Demographic features based decision tree model provided best results (77% accuracy), followed by a hybrid features’-based model also using decision tree (76% accuracy). The optimally selected features of the hybrid model were the academic board of degree (a domographic feature), two matriculation subjects (mathematics, and physics), one intermediate subject (chemistry), and marks in the entry test (conducted for the admission in graduate degree). The study shows models’ performance also depend on fine grained features (subject level marks) both in terminal degree (Intermediate) as well as in early education (Matriculation) which most of the universities don’t include in admission criteria. Interestingly, marks in Biology (relevant subject to doctor of veterinary medicine program under consideration) was not as much important as compared to marks in Mathematics &amp; Physics (Matriculation), and Chemistry (Intermediate). While demographic features, for ethical reasons, should not directly influence admissions decisions, they offer a treasure trove of information for targeted student support initiatives. We suggest to use decision tree and its variants for performance prediction and admission criteria mining due to their high performance and interpretability.</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

Bridging the gap: from traditional admissions to data-driven insights for predicting and supporting undergraduate performance

  • 期刊论文

  • 发表日期：

    2025年10月21日

查看全文
  • Muhammad Bilal,
  • Muhammad Omar,
  • Waheed Anwar,
  • Rahat H. Bokhari,
  • Gyu Sang Choi

摘要

This study leverages AI to predict undergraduate performance after first semester and unveil potent insights for refining admission criteria. We move beyond the siloed gaze of mere marks and entry tests, demonstrating that incorporating academic performance in specific subjects alongside demographic factors like gender, father’s profession, and exam board unlocks significantly improved early performance prediction. Demographic features based decision tree model provided best results (77% accuracy), followed by a hybrid features’-based model also using decision tree (76% accuracy). The optimally selected features of the hybrid model were the academic board of degree (a domographic feature), two matriculation subjects (mathematics, and physics), one intermediate subject (chemistry), and marks in the entry test (conducted for the admission in graduate degree). The study shows models’ performance also depend on fine grained features (subject level marks) both in terminal degree (Intermediate) as well as in early education (Matriculation) which most of the universities don’t include in admission criteria. Interestingly, marks in Biology (relevant subject to doctor of veterinary medicine program under consideration) was not as much important as compared to marks in Mathematics & Physics (Matriculation), and Chemistry (Intermediate). While demographic features, for ethical reasons, should not directly influence admissions decisions, they offer a treasure trove of information for targeted student support initiatives. We suggest to use decision tree and its variants for performance prediction and admission criteria mining due to their high performance and interpretability.