University dropout rates and suboptimal academic performance continue to pose significant challenges in higher education, frequently attributable to the absence of timely and personalized interventions. This study outlines a sophisticated framework that integrates predictive analytics with artificial intelligence to accurately identify students at risk of academic failure and deliver prompt support.

Data from 591 students enrolled in an Object-Oriented Programming course were used, including partial grades, demographic information, and Moodle activity logs. The Support Vector Machine (SVM) model was trained and validated using k-fold cross-validation, resulting in impressive performance: AUC 0.9830, 85% accuracy, 88% precision, 85% recall, and F1-score of 86%.

When the model detects a student at risk, a GPT-4o-mini-powered chatbot intervenes. This chatbot initiates personalized conversations using natural language processing and semantic search to guide students through course materials and suggest relevant learning activities. The chatbot achieved an 5.6-second average response time and a 100% success rate in task handling.

The results highlight the potential of the system as a scalable and proactive solution for academic intervention. By integrating predictive modeling with AI-driven tutoring, institutions can enhance learning personalization, improve academic performance and reduce dropout rates. This approach represents a promising advancement in the use of intelligent systems to promote student success in higher education.

Keywords: AI Chatbot, education, academic performance, risk detection.