Spatiotemporal Analysis of Road Crash Frequency and Severity in Dhaka, 2007-2021: Trend Detection, Change-Point Identification, and Corridor-Level Risk Persistence

Abstract

Road crash research in Dhaka has relied on cross-sectional severity models fitted to short observation windows, leaving the temporal behavior of the city's safety performance untested. This study analyses 17,497 police-reported crashes recorded in the Dhaka Metropolitan Area from 2007 to 2021, comprising 180 consecutive monthly observations, and combines non-parametric trend and change-point testing, interrupted time-series regression, a forecasting benchmark, and empirical-Bayes corridor ranking. Crash frequency and severity diverge. Annual counts rise significantly while the fatality share falls from 72.4% to 54.5%, yet annual fatal counts show no monotonic trend. The declining ratio therefore reflects a growing denominator rather than fatality reduction, and a 4.6-fold rise in recorded minor injuries indicates improving capture of less severe events. Structural breaks occur in the mid-2010s, preceding both the Road Transport Act 2018 and COVID-19 restrictions, whose effects prove statistically indistinguishable from the dominant secular trend. Twenty corridors carry 58.1% of all crashes, yet 59 of 64 corridors are indistinguishable from the city-wide fatality share of 64.2%, and concentration remains stable. Available covariates carry negligible severity information, with classification performing below the majority-class baseline. The findings identify a small, persistent set of treatable corridors and argue for expanding the crash record itself.