A Comparative Analysis of Machine Learning Techniques to Explore Factors Affecting Mathematics Success in Developing Countries: Turkey, Mexico, Thailand, And Bulgaria Case Studies

This study explores factors influencing mathematics achievement in Turkey, Bulgaria, Mexico, and Thailand using PISA 2018 data and machine learning models, comparing their performance. Both classification and regression models were utilized: linear regression, support vector machine, decision tree, and random forest for regression; logistic regression, support vector, decision tree, and random forest for classification. Additionally, XGBoost identified key predictors of math achievement, and K-Means filled missing data. According to results, key contributing factors across all countries included students' economic, social, and cultural status, study materials at home, sense of ownership, and family welfare. Regarding model success, random forests outperformed other models in both regression and classification, with Random Forest Regression achieving the highest R-square values (71%-84%) while linear regression has the lowest (22%-43%). In addition, the classification algorithms were analyzed in terms of binary and ternary classification, binary classification proved more successful than ternary, with RF accuracy scores ranging from 73% to 83% across countries. The study's findings offer valuable insights for selecting optimal algorithms for predicting math achievement, aiding decision-makers in enhancing educational outcomes.