Unified AI Framework for Real-Time Customer Churn Prediction Using Behavioral and Event-Log Anomaly Signals

Predicting churn age among customers is one of the most important challenges when working with a digital service platform; however, the majority of existing methods utilize mostly behavioral and transactional attributes and do not consider the influence of an anomalous event in a system on customer disengagement. Abnormal event-log patterns, including atypical access control and authentication failures, are likely to have a severe impact on user behavior and affection and frequently to churn. The paper introduces a single AI-driven real-time customer churn prediction framework combining behavioral and event-log generated anomaly signals. The proposed system uses unsupervised anomaly detection of system and security logs to extract behavioral deviations caused by events and models the resulting anomaly scores with behavioral features to predict churn using supervised classification. Experimentally, it has been demonstrated that the performance is significantly better than the behavior-only baselines with a churn detect probability of about 0.45 at 20 days, 0.68 at 10 days and 0.92 at 5 days before churn as compared to 0.75 with a baseline model. The structure also enhances the isolation of risks, and the median churn score is greater (0.69) as compared to the traditional methodologies (0.50). These findings indicate that event-log anomaly signals are quality early warning signs of customer churn.