Predicting school students' academic performance using ensemble machine learning techniques

Abstract

In schools, forecasting academic performance is essential, as it enables the earlier identification of those at risk, followed by targeted interventions. This study explores how machine learning (ML) models can be employed to forecast grades of school students from a given dataset with behavioral, academic, and demographic dimensions. There is a detailed methodology in the research, comprising preprocessing of the data, Chi-squared feature selection, and development of the prediction models by the k-Nearest Neighbors (kNN), Support Vector Machines (SVM), Multi-Layer Perceptron Classifier (MLPC), Logistic Regression (LR), and Decision Trees (DT). Finally, hard and soft voting classifier ensemble models were incorporated for enhancing the accuracy of predictions. 10-fold cross-validation was employed for model testing, and the highest accuracy (79.1%) was achieved by the soft voting classifier model with MLPC model. Furthermore, SHapley Additive exPlanations (SHAP) method was utilized to enhance model interpretability by providing insights into feature importance. Key findings indicate that behavioral characteristics such as class attendance and online resource usage are essential determinants of academic performance. The research concludes that schools and academic administrators can identify at-risk students improved with the assistance of ML-based grade prediction models.