PDPBoost: An XGBoost-Based Method for Predicting Programming Problem Difficulty

With the increasing demand for programming skills across various industries, Online Judge platforms have become essential for improving programming proficiency. However, selecting appropriately challenging problems is key to effective adaptive learning on these platforms. Existing methods for determining problem difficulty face limitations: traditional approaches are constrained by subjectivity and inefficiency, while current deep-learning-based models struggle with interpretability and require extensive training time. In this paper, we propose PDPBoost, a novel method utilizing XGBoost models to Predict the Difficulty of Programming problems. Our method combines dense and sparse feature extraction with ensemble learning to improve prediction accuracy, speed, and model interpretability. Experimental evaluations on two real-world datasets demonstrate the superiority of our approach in both predictive performance and computational efficiency.