A Geospatial and Machine Learning Approach to Identifying and Predicting High-Risk Segments along Major Highway Corridors in Northern Nigeria

Road transport is the backbone of Nigeria's economic connectivity, yet violent insecurity along northern highways has become a recurring hazard that undermines the resilience of critical transport infrastructure. Little data-driven research predicts these threats along specific corridors, so security planning is rarely informed by evidence about which stretches of road carry the greatest risk. This study frames highway insecurity as a spatial hazard and develops a geospatial and machine learning approach to identify and predict high-risk road segments along two contrasting routes: the high-traffic Abuja-Kaduna-Zaria corridor of about 266 km and the banditry-prone Kaduna-Birnin Gwari route of about 83 km. Geocoded incident records, road networks extracted by shortest-path routing, gridded population and land cover were integrated across 174 segments of 2 km. Because the risk label derives from incident records, predictors were computed from 2015 to 2021 events and the label from 2022 to 2025, so the model is judged on a period it has not seen. Hotspot analysis iden-tified 16 significant high-risk segments concentrated around Kaduna, while global spatial autocorrelation was not significant, indicating risk is concentrated in a few identifiable places rather than spread along the routes. An ensemble of five classifiers reached a cross-validated PR-AUC of 0.586 against a chance baseline of 0.057, though classification at the conventional threshold remained weak, indicating such models are better used to rank segments than to label them safe or unsafe. Past kidnapping was the strongest predictor. The resulting vulnerability map offers a proactive basis for resource allocation and early warning.