Confidence-Aware Academic Risk Prediction Using Federated Deep Learning: A Comparative Study with CatBoost and Machine Learning Models

This research paper proposes a trust-based frame work for predicting academic risk using the Federated Deep Learning algorithm, with a comparative evaluation against CatBoost and Deep Learning methods. The study aims to support the early detection of at-risk students by combining predictive accuracy with confidence-based decision-making. n order to accomplish this, various educational datasets were combined into a single analytical framework built on a data warehouse. Key indicators pertaining to student performance, engagement, support, and prior academic endeavors were then extracted through preprocessing, feature preparation, and exploratory analysis. Federated Deep Learning was adopted as the primary model due to its effectiveness and ability to model complex nonlinear relationships. Its performance was compared with Federal Learning and Deep Learning models to examine the predictive power across different models. The experimental results showed that the proposed Federated Deep Learning approach achieved an accuracy of 0.91, demonstrating strong performance in predicting academic risk. This study focuses on enhancing confidence awareness to improve the reliability of predictions. The results indicated that this proposed framework provides an effective and practical basis for building reliable early warning systems.//this paper is in publishing process and it will be published in sjr journal