Predicting Student Attrition in Kenyan Universities: A Comparative Analysis of Machine Learning Algorithms

Abstract

One of the primary goals of higher education institutions is to provide high-quality education and ensure a high completion rate. Reducing student attrition is one strategy for attaining high-quality education. Identifying students who are susceptible to dropping out and the variables that lead to dropouts are essential to achieving this. The purpose of this research was to ascertain how machine learning models might be used to forecast student attrition in Kenyan universities. Based on a number of classification criteria, such as F1 score, precision and accuracy, the study assessed and contrasted the performance of numerous algorithms, including Decision Trees, Random Forest, Naive Bayes, and Logistic Regression. The analysis demonstrated how well Logistic Regression worked, outperforming the other models and consistently striking a balance between precision and recall. Decision Trees and Random Forest, despite showing improvements through hyperparameter tuning, still struggled to identify students at risk of attrition. Naive Bayes, while relatively balanced, did not match the performance of Logistic Regression. The study provided a comprehensive overview of each model's strengths and limitations and suggests future work to further optimize the models for better predictive performance.