Predicting Student Examination Performance to Improve Remedial Programs

Education is a critical pillar of success in life, and every child should have the right to quality education. As such, administrators should make every effort to provide an environment that fosters student success at every level. A crucial assessment period for children in St. Vincent and the Grenadines occurs at the Grade 6 level where students sit the Caribbean Primary Exit Assessment. In this study, we propose several models to predict student performance in Mathematics using diagnostic examination results for Math and English at Grade 2 and Grade 4 levels. We created additional features via a greedy approach to factor in the interactions of the provided features. The models used for prediction include the XGBRegressor, Principal Component Regression, KNN Regression, Lasso Regression, and a DNN. We compared these models using the metrics of Root Mean Square Error (RMSE), and the Co-efficient of Determination (R
²
). We also performed a number of experiments applying two scalers, StandardScaler and QuantileTransformer. We obtained the best performance using the XGBRegressor model with an RMSE, MAE and R2 of 9.71%,7.59% and 61.56% respectively.