A neural network regression model for predicting student learning success based on prior achievements

Siberian State University of Science and Technology named after M.F. Reshetnev, Krasnoyarsk, Russia

^* Corresponding author: dorrer_mg@sibsau.ru

Abstract

The paper describes a project utilizing data analysis tools to predict student performance based on their prior achievements. The task was addressed using historical educational data from over 35,000 students over a span of seven years, containing information on 1.24 million grades. Neural network regression tools were employed to build models that predict future grades, thereby enhancing educational processes. The predictive capability of the model was assessed using the coefficient of determination and the root mean square error (RMSE) through 10-fold cross-validation of the dataset into training and testing sets. More than 70% of the developed grade prediction models demonstrated a coefficient of determination greater than 0.7, with the RMSE of predicted grades from actual values being less than one point on a five-point scale. This indicates a satisfactory solution to the prediction problem.