Predicting Students Academic Performance using Ridge, LASSO and Elastic Net with Efficient Model Selection: A Case Study of Federal University Gusau Nigeria

Abstract

Academic establishments function in a very demanding and competitive environment. Most of the schools’ face challenges in providing the students with a high-quality education, creating methods for assessing student achievement, analysing performance, and anticipating the needs of their students in the future. This research proposes a sparse regression model that solves the problem of multicollinearity among the six variables. The significant variables that determined the academic performance at graduation were identified, using ridge, LASSO and elastic net. Evaluation metrics and seven selection criteria were used to select the best model. The results show that the age of the student has no impact on the CGPA at graduation. For the training data, the elastic net using SGMASQ significantly suggests better performance in comparison to other models. For the testing data, the elastic net using SGMASQ significantly shows better results in comparison to other models including the training data.