Exploring Predictive Models for Students' Performance in Exams: A Comparative Analysis of Regression Algorithms

Abstract

Student-centered analysis of academic performance is also the most important aspect in improving education by being able to determine what measures work best, individualized learning approaches, and intervention programs. In this study, we performed a detailed analysis based on the Students Performance in Exams dataset and different regression methods to estimate students' grades. We sought to assess the functioning of numerous metrics and determine an optimal model for this task. Our descriptive analysis identified meaningful trends within this dataset, as it includes central factors like 'gender, raceethnic diversity-based status of a student,' and parental education level based on which the children are catered to by informing them about important lunches and test preparation courses alongside scores in Math, Readings, Writing etc. We used a wide range of regression models: XGBoost, CatBoost, GradientBoostingRegressor, etc. Metrics like Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Average Marginal Loss were used to assess each model rigorously. Importantly, the XGBoost model gave out an MSE value of 0.028, which was the best among all values obtained from various other models. The superiority of the XGBoost model is supported by the excellent performance that was reported across many metrics. This work can be important for informing educational practitioners and policymakers regarding the best possible accurate and realistic model that would predict the students' outcome results. Educational data analytics incorporating the XGBoost model can be used for the customization of interventions and mapping resource allocation while promoting a results-oriented approach based on data in education. This study is a step towards the accumulation of knowledge on educational data analytics. It can serve as a background for further research aimed at improving predictive models regarding student performance.