Scholarlink

<p>One of the significant challenges instructors face in education is identifying students who may be at risk of academic failure. Forecasting these students early is essential as it allows instructors to suggest timely interventions to improve student outcomes and avoid failure. Nevertheless, it is not a straightforward task due to the diverse academic levels of students and the varied assessment activities within different courses. In this context, we propose the SMOTE + GA + GRU model to predict at-risk students, which consists of four key steps: data preprocessing to clean and balance the data using SMOTE to address imbalanced classes and ensure the model is trained on more representative data; and GA and GRU to learn from the data. Then, we use a genetic algorithm to select informative features for student performance prediction. After that, a training step is performed using GRU with 128 Units. The final step focuses on evaluating the model’s efficiency. The findings confirm the proposed model’s efficiency, outperforming existing approaches with an accuracy of 90%. The proposed model can be helpful in educational environments, assisting instructors in swiftly identifying students at risk.</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

On the Use of Advanced Learning Analytics in Student At-Risk Prediction: A Deep Learning Model

  • 期刊论文

  • 发表日期：

    2026年2月10日

查看全文
  • Nesrine Mansouri,
  • Mourad Abed,
  • Makram Soui,
  • Mohammed Kutbi

摘要

One of the significant challenges instructors face in education is identifying students who may be at risk of academic failure. Forecasting these students early is essential as it allows instructors to suggest timely interventions to improve student outcomes and avoid failure. Nevertheless, it is not a straightforward task due to the diverse academic levels of students and the varied assessment activities within different courses. In this context, we propose the SMOTE + GA + GRU model to predict at-risk students, which consists of four key steps: data preprocessing to clean and balance the data using SMOTE to address imbalanced classes and ensure the model is trained on more representative data; and GA and GRU to learn from the data. Then, we use a genetic algorithm to select informative features for student performance prediction. After that, a training step is performed using GRU with 128 Units. The final step focuses on evaluating the model’s efficiency. The findings confirm the proposed model’s efficiency, outperforming existing approaches with an accuracy of 90%. The proposed model can be helpful in educational environments, assisting instructors in swiftly identifying students at risk.