Wildlife-vehicle collisions in the Mid-Zambezi Valley: Driver perceptions and species-specific risk patterns

• Emblematic species in the Mid-Zambezi Valley face elevated wildlife-vehicle collision risks, with critical corridors identified to guide targeted conservation measures and enhance road safety for both wildlife and local communities. • Integrating driver perception surveys with empirical collision data highlights the influence of socio-demographic factors, such as driver age and experience, on collision frequency, informing tailored human-wildlife conflict mitigation strategies. • Unsupervised machine learning techniques (K-means and hierarchical clustering) effectively classify species and temporal collision risk groups, enabling evidence-based, species-specific mitigation planning across heterogeneous savannah landscapes. Wildlife-vehicle collisions (WVCs) are an escalating threat in the Mid-Zambezi Valley, a critical transboundary wildlife corridor where expanding transport infrastructure increasingly intersects major movement routes of ecologically important species. However, limited empirical work has combined driver-based data with advanced machine learning techniques to characterize species-specific collision risks in this region. This study surveyed 100 truck drivers operating along the A1 highway and analysed collision records using K-means and Hierarchical Agglomerative Clustering (HAC) to identify functional species groups and spatial–temporal risk patterns. Results show that 73 % of collisions occur at night, largely involving nocturnal or crepuscular species such as hyenas, civets, and warthogs, while high-impact megafauna including elephants, lions, and buffalo form a distinct cluster associated with severe safety risks and predictable movement corridors. Driver age and cumulative experience were significantly associated with collision exposure, although perceptions of WVC causes diverged from empirical patterns, highlighting gaps in road safety awareness. The integration of unsupervised machine learning with ecological interpretation represents an advance over previous studies by enabling data-driven, species-specific risk profiling rather than generalized hotspot mapping. These findings provide new knowledge on functional WVC groupings in African savannah systems and offer evidence-based guidance for mitigation, including targeted nighttime traffic management, vegetation clearance at key points, and corridor-sensitive infrastructure such as wildlife crossings. By informing conservation planning, road safety policy, and regional connectivity management, the study contributes to continental priorities under the African Union’s Agenda 2063 and supports SDG 15.7 on reducing threats to biodiversity.