Concept drift, a phenomenon where statistical properties of variables change over time, poses significant challenges to predictive models by reducing their performance in dynamic environments. This work presents a framework for detecting concept drift in Bayesian Networks by leveraging probabilistic inference and dynamic updates of Conditional Probability Tables (CPTs). Drift detection is achieved using sliding windows to monitor deviations in probabilistic inference, while CPT updates are guided by adaptive retraining to realign to the evolving data distribution. The proposed approach is quantitatively evaluated using a symmetric divergence metric, ensuring robust tracking of changes across data segments and adaptability to diverse drift scenarios. It was evaluated on synthetic datasets with induced drift, showing excellent performance in drift detection and parameters’ update.