Predicting Student Performance to Improve Academic Advising Using the Random Forest Algorithm

Introduction

Student success is one of the main goals of educational institutions. It is measured by academic performance, or the extent to which students meet the standards defined by these institutions. According to Tuckman (1975), student performance refers to observing student knowledge, skills, concepts, understanding, and ideas. In the higher education environment, academic student performance depends on instructors’ and program coordinators’ standards and reflects the achievement of their short- and long-term educational goals (Sundar, 2013).

Prediction of accurate academic student performance in universities is considered an important tool that helps in various decisions related to student admission, retention, graduation, and adapted educational support based on student data observation. Student performance is a significant indicator in measuring institutions’ effectiveness and a crucial factor in students’ future success, particularly in countries’ prosperity. For this reason, higher education institutions focus today on improving student performance and enhancing the quality of their educational programs. An in-depth analysis of the learners’ previous records can play a vital role in providing quality education to learners.

Early prediction of academic student performance helps institutions provide appropriate actions to improve students’ retention and success rates. Educational data mining (EDM) involves analysis and improvement in the prediction methods of student performance. With EDM techniques, researchers can develop prediction models to detect, monitor, and improve student achievement (Alyahyan & Düştegör, 2020).

Predicting academic student performance may also improve curriculum content and plan for adequate academic advising for students. Data mining techniques allow researchers to examine data sets and obtain conclusions that help improve the educational learning process. Various techniques have been applied for this purpose. Machine learning, collaborative filtering, Bayesian networks, artificial neural networks, random forest decision trees, rule-based systems, and correlation analysis have been applied to predict the risk of dropping out of the university, students’ achievement, or grades. All these techniques classify the significant factors that affect and foresee overall student academic performance. However, they differ in precision/accuracy, complexity, and sample data size requirements.

This paper focuses on developing a prediction model of students’ academic performance based on their high school average score and second and third-year grades in a four-year information technology program. It explores the performance of random forest (RF) machine learning in predicting student performance to achieve high predicting accuracy. The proposed methodology proved its worth by achieving accurate results. The result can be used by students, advisors, and program coordinators to reduce education difficulties, improve students’ results, provide better quality education, and develop plans for education policy.