Analysis of Learning Patterns and Prediction of Students’ Academic Performance in Virtual Education Using Data Mining Algorithms

Abstract

The present study aimed to identify students’ learning patterns in virtual education environments and predict their academic performance using data mining and machine learning algorithms. This study employed a data-driven approach using educational data mining techniques. The statistical population consisted of secondary school students in Semnan Province, Iran, from whom data related to 400 students actively participating in virtual learning environments were collected. Data were extracted from online learning management systems, school databases, and electronic questionnaires and included variables such as self-regulated learning, digital skills, family support, teacher feedback quality, academic motivation, and prior academic performance. After preprocessing procedures including data cleaning, normalization, and removal of incomplete records, the K-Means algorithm was used for clustering learning patterns. In addition, Random Forest, XGBoost, MLP neural network, Decision Tree, SVM, KNN, Logistic Regression, and Naive Bayes algorithms were applied to predict academic performance. Model performance was evaluated using Accuracy, Recall, Precision, and F1-Score indices. The clustering analysis identified three major learning patterns: “independent and successful learners,” “at-risk learners,” and “learners with hidden potential.” The first cluster demonstrated high levels of self-regulation, digital competence, and family support, whereas the second cluster was characterized by weak self-regulation skills and inadequate learning conditions. Among predictive models, the Random Forest algorithm achieved the best performance with 92% accuracy, 92.8% recall, and an F1-score of 91%. XGBoost ranked second, followed by the MLP neural network. Feature importance analysis revealed that self-regulated learning, teacher feedback quality, and family support were the most influential predictors of students’ academic performance. The findings demonstrated that students’ learning patterns in virtual education environments are highly heterogeneous and influenced by a combination of cognitive, motivational, interactive, and environmental factors. Furthermore, data mining algorithms, particularly Random Forest, showed high capability in analyzing educational data and predicting academic performance. These approaches can facilitate personalized learning design and intelligent decision-making in virtual education systems.