Predicting Students’ Mathematics Scores from Reading Scores Using Supervised Learning

Abstract

This study aims to predict students’ mathematics scores based on their reading scores using a supervised learning approach. The dataset used is from Students' Performance in Exams (Kaggle), consisting of 1,000 student records, and was analyzed using Microsoft Excel and Google Colaboratory. The data was divided into training and test data with a ratio of 80:20. The research stages included descriptive statistical analysis, data visualization, Pearson correlation testing, linear regression model development, and model performance evaluation using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and coefficient of determination (R²).  Prior to modeling, regression assumptions including linearity, normality of residuals, and homoscedasticity were examined to ensure model validity. The results showed a strong positive relationship between reading and math scores with a correlation coefficient of 0.818. The linear regression model produced an MAE of 7.281, an RMSE of 8.818, and an R² of 0.680. Decision Tree Regressor was selected as a comparison model because it represents a non-linear and non-parametric supervised learning approach commonly used in educational data mining. This study contributes to educational data mining literature by demonstrating that interpretable regression models explain significant mathematics achievement variance, rivaling the performance of non-linear alternatives.