A MACHINE LEARNING FRAMEWORK FOR STUDENT PERFORMANCE FORECASTING: BEST FIRST SEARCH WITH LOGISTIC REGRESSION AND DECISION TABLE IN EDUCATIONAL DATA MINING

Abstract

Predicting student academic performance is a critical priority for educational institutions striving to enhance learning outcomes and provide timely support for students at risk of failure. Machine learning offers robust techniques for analyzing student data and identifying influential factors that shape academic achievement. This study conducts a comparative evaluation of two machine learning classification algorithms Logistic Regression (LR) and Decision Table (DT) to predict students’ final grades (G3). Using the Students’ Academic Performance dataset from the UCI Machine Learning Repository, which includes 395 records and 33 attributes, the research employs the Best-First Search (BFS) method for feature selection. BFS effectively reduces the dataset to 13 key attributes, encompassing demographic details, personal factors, and prior academic performance indicators (G1 and G2). Both LR and DT models were developed and assessed using evaluation metrics such as correlation coefficient, Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). The results demonstrate that the Logistic Regression model outperforms in terms of RMSE, showing a stronger correlation and lower error rates. Overall, the study contributes to the field of Educational Data Mining (EDM) by providing insights into predictive modeling approaches that can support early identification and intervention for students requiring academic assistance.

Keywords: Educational Data Mining, Machine Learning, Best First Search, Decision Table, Logistic Regression.