Educational data mining has gained prominence as institutions seek to improve student outcomes through predictive analytics. This research examines the effectiveness of machine learning models in predicting academic performance across different cultural and educational contexts. We analyzed data from 2,847 students across 12 high schools in California, Singapore, and Indonesia, implementing random forest, support vector machine, and neural network algorithms. Our cross-cultural analysis reveals that while socioeconomic factors remain significant predictors globally, cultural variables such as collectivism scores and educational system structures significantly influence model accuracy. The study provides insights for developing culturally-sensitive educational interventions.