Predicting Student Academic Performance: A Machine Learning Approach and Feature Analysis

Predicting student academic performance is a challenging task and, at the same time, has significant implications for educators and policymakers in the field of education. By utilizing machine learning techniques, this article seeks to explore the relationship between various features across six categories: demographic factors, personality traits, skills, favorite activities, relationships with others, out-of-school activities on one hand, and academic performance in terms of Grade Point Average, on the other. The data utilized in this study has been collected through several surveys conducted in one of the schools in Iran over multiple years and educational levels, which form the basis of the analysis. Using CRISP-DM methodology, a predictive model is developed based on CatBoost Regressor. A predictive model with an R-squared value of ۰.۸۷ is developed. Moreover, the analysis of feature importance reveals that positive personality traits such as "Interest in studying," "The quality of homework," "Contentment," "Self-regulation," and "Logical thinking and reasoning" skills are among the most predictive features affecting students' academic performance which is rooted in and supported by some of the well-known psychological theories such as Self-Determination Theory. The contribution of the current research includes the development of a highly accurate prediction model based on the machine learning approach to predict student academic performance in terms of their GPA and to extract the most important features that influence it. This study is unique in this field due to the incorporation of various features and data collection across different years and educational stages.

کلیدواژه ها:

Educational Data Mining (EDM) ، Machine learning ، Academic Performance ، intrinsic motivation ، regression algorithms ، Self-Regulation