This study proposes an interpretable machine learning framework using eXplainable AI techniques, specifically LIME and SHAP, to predict and explain code submission outcomes in an online judge system. The framework utilizes the Random Forest, XGBoost, and LightGBM models, achieving a top prediction accuracy of 91.23% with XGBoost. The identified influential factors include memory usage, CPU time, code size, and completion rate. Successful submissions were characterized by efficient memory usage, low CPU time, concise code size, and high completion rates of 80% or above. The integration of LIME and SHAP provides a comprehensive and interpretable approach for optimizing code submissions in online judge systems. This study underscores the importance of interpretability in machine learning applications in educational and software development contexts, offering actionable insights to support users of online judge systems.