A Hybrid Deep Learning Model To Predict High-Risk Students In Virtual Learning Environment

Abstract

The rapid growth of virtual learning environments (VLEs) has increased the need for intelligent systems that can identify students who are at risk of poor academic performance or disengagement. This study proposes a hybrid deep learning–based predictive framework for detecting high-risk students by analyzing behavioral and engagement data collected from online learning platforms. The system is implemented as a web-based application using a Flask framework, where student interaction data such as time spent weekly, quiz score average, forum participation, video watching percentage, assignment submission rate, login frequency, session duration, device type, course difficulty, and region are utilized as predictive features. Data preprocessing includes missing value removal, label encoding for categorical attributes, and feature normalization using a standard scaler. The dataset is then divided into training and testing subsets to evaluate model performance. Multiple machine learning and deep learning models—including Convolutional Neural Networks (CNN), XGBoost, Random Forest, and Decision Tree classifiers—are trained and compared to determine the most effective approach for predicting student engagement risk levels. The system classifies students into low, medium, or high engagement risk categories, enabling early identification of learners who may require academic support. Experimental results demonstrate that integrating deep learning with traditional machine learning models enhances prediction accuracy and provides a scalable solution for improving student retention and personalized learning interventions in virtual education systems.