Academic burnout in university students constitutes a public mental health problem, with important repercussions on the academic performance and psychological well-being of students. Traditional assessment methods are predominantly reactive and lack predictive capacity for early intervention. This study proposes a machine learning-based predictive model using Extreme Gradient Boosting (XGBoost) to identify students at risk for academic burnout. The model was initially trained with survey data from 71 Peruvian university students, considering 18 academic, emotional and contextual variables. To improve robustness and generalizability, synthetic data were generated using the CTGAN model from the Synthetic Data Vault (SDV) library, expanding the sample to 300 instances. The Oldenburg Burnout Inventory for Students (OLBI-S) instrument was used as the target variable. The model achieved an average R2 of 0.70 (95% CI: 0.59–0.82) and an RMSE of 5.04 (95% CI: 4.71–5.36) under five-fold cross-validation. When evaluated on a hold-out set, performance increased to R2 = 0.84 and RMSE = 4.43, indicating strong generalization. The SHAP (SHapley Additive exPlanations) framework was used to ensure the interpretability of the model. Finally, the model was deployed as a REST API service using FastAPI, enabling real-time risk assessments. This approach enables early detection and promotes proactive mental health strategies within the university setting.