Comparative Study of Machine Learning Models for Student Performance prediction

Abstract

Predicting how well students will perform is becoming an increasingly useful application of machine learning in education. Schools, colleges, and universities collect different types of student information, such as attendance, previous grades, study habits, participation, and academic activities. When this information is analyzed properly, it can help teachers and institutions recognize students who may be struggling and provide support before their performance drops further. Traditional approaches to evaluating students usually rely on examinations, teacher observations, assignments, attendance records, and previous academic results. These methods are valuable, but they may not always capture the relationships between several factors at the same time. For example, a student may have reasonable examination results but still be at risk because of declining attendance or reduced participation. Machine learning can help by finding patterns in historical student data and using those patterns to estimate future academic performance. This research presents a comparative study of four machine learning models: Linear Regression, Decision Tree, Random Forest, and Support Vector Regression (SVR). The proposed methodology includes data collection, data preprocessing, exploratory analysis, feature selection, model training, and performance evaluation. Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-squared (R²) are used as evaluation measures. The study also discusses the strengths, limitations, and practical uses of each model. Factors such as previous grades, attendance, study time, and assignment performance are considered particularly relevant. Overall, machine learning can provide useful information about student performance and can support educational institutions in taking early and more targeted action when students appear to be at risk.