Search for: All records

Predicting students' performance early in programming courses is crucial because it allows instructors to intervene early, improving learning outcomes. Currently, no existing platforms can effectively forecast student performance in programming activities based on students' developed code. Forecasting student scores based on their programming activities is challenging because the accuracy of different predictive models often varies throughout these activities. To address this challenge, we introduce a novel framework utilizing Mixture of Experts (MoE). The MoE method combines insights from various neural networks and dynamically picks the most accurate predictions. This system significantly enhances the reliability of forecasting each student's performance within the first 15 minutes of a 30-minute programming session. By enabling early predictions, the MoE provides instructors with a powerful mechanism to understand and support the student learning process in real-time.