Impact of land use on bus driver inattentiveness: Insight into crash risks

This study examines how roadside land-uses, namely commercial, residential, educational, and recreational establishments located on a typical Indian Highway, i.e., National Highway, influence the temporally distributed bus driver inattention, such as distraction, drowsiness, and falling asleep behaviors and their associated influence on the crash severity occurring on the entire study corridor.

In this context, AI-powered Driver Monitoring Systems (DMS) were installed in a public bus fleet operated by 33 professional drivers over three months and generated multiple inattention events. These events were categorized into morning, afternoon, evening, and night periods. The above-mentioned roadside land-use features were extracted using high-resolution Google Earth imagery (0.5-1.0 m) by considering within a 100 m roadside buffer and validated through field surveys. Three years of inattention-related road crash data (2020-2022) were integrated. Analytical methods included descriptive statistics, structural equation modeling, correlation analysis, Generalized Linear Models (GLM) and Multiple Linear Regression (MLR) regression, and Artificial Neural Networks (ANN) modeling to estimate the associations among inattention events, land-use establishments, and crash risks. Model validity and robustness were assessed through sensitivity measures Receiver Operating Characteristic (ROC)-Area Under the Curve (AUC) metrics, and normalized importance variance values (NIVIA). ANN-based risk outputs were further visualized using spatial hotspots for crashes, fatalities, and injury severities. Spatial dependency and clustering were evaluated using Global Moran's I and Local Getis-Ord Gi* statistics.

Inattention behaviors exhibited strong spatial dependence (Moran's I = 0.32-0.48, p < 0.001), whereas crash counts themselves were spatially random. Recreational, residential, and commercial land-use segments significantly increased distraction and asleep-related inattention. Morning distraction emerged as a strong predictor of crash occurrence, while nighttime asleep events showed the highest association with fatal and major-injury outcomes. ANN-based risk surfaces and Gi* clustering consistently identified intensive daytime distraction hotspots in activity-dense land-use clusters and severe nighttime hotspots driven by falling asleep. Sensitivity was highest for nighttime asleep (84.84%) and morning distraction (70.04%), with MLR models achieving strong discrimination (ROC-AUC = 0.94 to 1.00), but they have less confidence in test data. The predominantly non-linear crash risk relationships were better captured by ANN models, as reflected by synaptic weight patterns and variable importance scores.

Driver inattention along the NH-65 corridor is shaped by both spatial context and circadian timing. Integrated evidence from statistical analyses, ANN risk estimation, and spatial hotspots indicates that daytime distraction and nighttime fatigue are the dominant contributors to crash severity. These findings highlight the need for time-specific and location-targeted interventions to enhance safety on this critical highway.