Abstract
This research aims to develop predictive models to identify road traffic accident risk
zones in Bangkok using Machine Learning and Deep Learning techniques. Accident
data from the Royal Thai Police (2020–2024) comprising over 70,000 cases were
analyzed through descriptive and predictive modeling approaches.
Random Forest, XGBoost, and Logistic Regression algorithms were employed with
temporal, spatial, demographic, and environmental variables. The results indicate
that the midnight to early morning period (00:00–06:00) posed the highest risk,
with motorcycle riders aged 20–25 years being the most vulnerable group. High-risk
districts included Lat Krabang, Bang Khun Thian, Prawet, Don Mueang, and Min
Buri, often located in outer Bangkok areas with complex road networks and logistics
routes.
The Random Forest model achieved the best predictive performance (ROC-AUC =
0.76, Balanced Accuracy = 0.72). Spatial analysis further revealed high–high clusters
of fatal crashes around major intersections, schools, and convenience stores.
Policy recommendations include establishing Road Safety Zones, improving street
lighting and traffic signage, developing a real-time crash risk monitoring system,
and creating a National Crash Data Committee to integrate road safety data across
agencies.
