Selecting a topic orientation is a crucial step for undergraduate students to finish their university coursework. However, they always struggle to make a decision since they doubt themselves and don’t fully comprehend their strengths. The division between research-oriented and application-oriented routes poses a challenging choice point in the context of the International University-Vietnam National University. It affects the optimization of institutional resources as well as individual academic success. Though there is a significant methodological gap in the systematic use of these tools for thesis orientation prediction, contemporary educational data mining methodologies have shown effectiveness in a variety of academic decision-making scenarios. With no factual underpinnings for orientation optimization, current selection paradigms mostly depend on informal discussions and subjective evaluations. Based on students’ academic performance criteria, this study creates a machine learning-based system to forecast the best thesis orientations. Using Random Forest, XGBoost, Decision Tree, and Support Vector Machine classifiers on extensive academic datasets, the study investigates relationships between academic trends and thesis results. Technical and theoretical coursework have varied predictive values, and the results show statistically significant correlations between certain academic performance patterns and thesis success indicators after thorough data pretreatment and cross-validation processes. The resulting approach contributes to educational data mining, advances knowledge of the connection between research aptitude and academic success, and offers significant promise for evidence-based thesis advice.