Deep Learning–Driven Crime-Aware Multi-Factor Route Safety Prediction Using Real-Time Environmental and Contextual Data

Traveler safety is a critical concern in modern Intelligent Transportation Systems (ITS), where conventional ‎navigation algorithms primarily optimize for distance or travel time while overlooking contextual safety factors. ‎This study introduces a Deep Learning–Driven Crime-Aware Route Safety Prediction Framework that fuses ‎static, dynamic, and crime-based contextual attributes for segment-level risk assessment. The model extends ‎earlier machine-learning frameworks by incorporating a Deep Neural Network (DNN) capable of learning ‎complex nonlinear relationships among environmental, demographic, and crime-contextual variables. Real-‎time data from Google Maps, OpenStreetMap, OpenWeatherMap, and TomTom APIs are combined with ‎synthetically generated crime-risk indicators derived from spatial density, place-type exposure, and temporal ‎crime propensity. Comparative evaluation of Decision Tree, Gradient Boosting, Random Forest, and DNN ‎models demonstrates that the DNN achieves the highest predictive accuracy (R² = 0.9987, ROC–AUC = 0.999) ‎while maintaining robust classification reliability. All models consistently identified Route 3 as the safest and ‎Route 1 as the most risk-prone corridor. By combining the interpretability of Gradient Boosting with the deep ‎learning capacity of DNNs, the proposed framework enhances route-level safety prediction and scalability for ‎real-time, human-centric ITS applications‎.