A Hybrid K-Means and Random Forest Framework for Crime-Aware Route Optimization in Urban Navigation Systems

With increasing urban crime rates, risk-free navigation has emerged as a prime issue in intelligent transportation systems and smart city planning. In this paper, we introduce a hybrid computational model for risk-aware route optimization using the combination of spatial clustering and ensemble learning over geo-temporal crime data. The core methodology employs K-Means clustering to detect and visualize crime hotspots based on crime intensity, type, and spatial distribution. These clusters guide the development of weighted crime density map penalizing routes through high-risk areas, including Random Forest classifier trained on featureengineered geospatial crime data with parameters like time of day, crime type, location frequency, and historical severity indices. This model calculates the relative danger score of unseen or new areas, and provides adaptive route evaluation in real time. Each candidate route, retrieved through OpenRouteService API, is expressed as a waypoint sequence and assessed through a multi-factor safety index. This index considers both crime density and a severity-aware score calculated through a normalized version of the Morile classification system and legal crime weights. For further refinement, a severity-weighted safety score is calculated through a customized mathematical model considering proximity to crime clusters and incident types. Experimental evaluation over UK crime datasets shows a considerable reduction in route risk exposure at the expense of comparable travel distance and time to traditional shortest-path algorithms. This research demonstrates the viability of combining clustering, supervised learning, and geospatial modeling for real-time, crime-aware navigation and presents a generalized framework suitable for various urban settings worldwide.