Previous TIMSS studies have employed conventional statistical methods, focusing on selected few indicators. The purpose of this study was to explore and identify important variables to predict students mathematics achievement, utilizing as many student, teacher, and school variables as possible via random forests, a popular machine learning technique. TIMSS 2015 Korean 8th graders student, teacher, and school datasets were merged to extract important predictors for students mathematics achievement. The prediction accuracy, sensitivity, and specificity of the model were 78%, 83%, and 73%, respectively. Among 413 TIMSS variables explored, variables identified as having the highest variable importance were all student variables, consistent with previous research. Scientific importance of the study was discussed as well as further research topics.