A comparative study of machine learning algorithms for prediction of students performance

Abstract

In today's evolving educational landscape, predicting student performance is crucial for resource allocation and support. While universities have explored this area, middle and high schools remain under-researched. This study aims to fill this gap through a literature review, emphasizing data mining techniques. Analysis of factors like gender, parental education, lunch type, and test preparation enabled accurate predictions. These insights empower institutions to make informed decisions, offer targeted support, and enhance student outcomes. Machine learning models can analyze large amounts of data, including data about students' past performance, demographic information, and other relevant factors. By applying a variety of algorithms, researchers and educators can create predictive models that outperform traditional methods. These models can configure educational content and make modifications to meet individual student needs. By continuously analyzing student performance and learning patterns, machine-learning models can recommend personalized learning plans, resources, and assignments. This approach enhances the learning experience and helps students understand and recollect information better. The study of comparing machine learning algorithms to predict student performance has a transformative impact on education. Various machine algorithms, such as Random Forest, SVM, Decision Tree, XGBoost, CatBoost, LightGBM, and Naive Bayes, were explored and assessed using R scores to predict the students’ performance and compare the results of these models. A comprehensive exploration of machine learning techniques applied to forecast student academic achievements is undertaken. The study rigorously evaluates various algorithms, such as Decision-Trees, support vector machines (SVM), and naive bayes, with meticulous attention to feature selection and hyperparameter tuning. Notably, Decision Tree and SVM models emerge as frontrunners, demonstrating impressive accuracies of 86.12% and 96.00%, respectively, while the Naive Bayes model also exhibits competitive performance, showcasing its potential in predicting student success. The research underscores the value of machine learning in educational contexts and offers actionable insights for institutions aiming to enhance student support and intervention strategies.