BER-PEF: Unified Human Mobility Predictability Evaluation via Bayes Error Rate Estimation
The BER-PEF framework is proposed to transform Bayesian error rate estimation into an observable protocol for evaluating迁移 predictive capabilities. This framework maps symbolic sequences, numerical trajectories, and contextual features into a unified feature-label space. The performance of the estimator is evaluated by measuring the deviation from the shared reference interval through controlled perturbation curves. Experiments on datasets from Foursquare NYC, TKY, GeoLife, and T-Drive show that BER-based estimators exhibit lower reference differences compared to existing methods for symbolic sequences and numerical trajectories, and can track changes in predictive performance under perturbation; comprehensive analysis indicates that aggregating multi-level perturbation evidence is more reliable than a single unperturbed observation for assisting in estimator selection. BER-PEF provides a unified and verifiable approach for evaluating predictive capability estimators for heterogeneous mobile data in the absence of true migration predictive capabilities.