Scholarlink

This study uses machine learning models to predict higher education student success and dropout rates based on academic performance, demographics, and socio-economic factors. The research evaluates three machine learning algorithms—Random Forest Regression, Logistic Regression, and AdaBoost Regression—on a dataset of 4424 student records. Among these, Logistic Regression proved to be the most reliable model, achieving the highest accuracy (0.918) and offering easy interpretability. Random Forest provided valuable insights into the importance of different features, while AdaBoost demonstrated strength in handling imbalanced datasets. The study found that academic performance indicators, such as approved curricular units and grades, are the most important factors in predicting student outcomes. However, challenges like imbalanced data and reliance on static features limited the scope of predictions. The research highlights the potential for improvement by integrating real-time data and exploring more advanced machine learning models. These findings serve as a stepping stone for higher education institutions to adopt data-driven approaches to enhance student retention and success.

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

Machine Learning Models for Early Prediction of Student Success and Dropout in Higher Education

  • 章节

  • 发表日期：

    2026年

查看全文
  • Nyoman Ayu Gita Gayatri,
  • Yaya Heryadi,
  • Ilvico Sonata,
  • Abba Suganda Girsang,
  • Lili Ayu Wulandari

摘要

This study uses machine learning models to predict higher education student success and dropout rates based on academic performance, demographics, and socio-economic factors. The research evaluates three machine learning algorithms—Random Forest Regression, Logistic Regression, and AdaBoost Regression—on a dataset of 4424 student records. Among these, Logistic Regression proved to be the most reliable model, achieving the highest accuracy (0.918) and offering easy interpretability. Random Forest provided valuable insights into the importance of different features, while AdaBoost demonstrated strength in handling imbalanced datasets. The study found that academic performance indicators, such as approved curricular units and grades, are the most important factors in predicting student outcomes. However, challenges like imbalanced data and reliance on static features limited the scope of predictions. The research highlights the potential for improvement by integrating real-time data and exploring more advanced machine learning models. These findings serve as a stepping stone for higher education institutions to adopt data-driven approaches to enhance student retention and success.