Machine learning and Geographic Information Systems (GIS)-based model for road crash severity prediction and...

Area of Science:

• Public Health
• Data Science
• Transportation Engineering

Background:

• Road traffic injuries are a significant public health issue in low- and middle-income countries.
• Jordan faces challenges with road safety, necessitating advanced analytical approaches.

Purpose of the Study:

• To develop a hybrid machine learning (ML) and Geographic Information Systems (GIS) framework for predicting road crash severity.
• To identify behavioral, environmental, and infrastructural factors contributing to injury outcomes in Jordan.

Main Methods:

• Analysis of 11,345 crashes using a two-stage approach: association rule mining and ML models (Decision Tree, Random Forest, AdaBoost).
• GIS-based kernel density estimation to identify spatial hotspots of severe and fatal crashes.
• Data preprocessing included encoding, imputation, normalization, and outlier treatment.

Main Results:

• Association rules identified links between nighttime driving, young drivers, and non-seatbelt use with fatal outcomes.
• GIS analysis pinpointed high-risk crash clusters in Zarqa, Awajan, and Al-Rusayfah.
• The Random Forest model demonstrated high predictive accuracy (98.5-99.9%), with key predictors including crash type, speed, time of day, driver age, and road characteristics.

Conclusions:

• The hybrid ML-GIS framework provides accurate crash severity prediction and reveals critical spatial and behavioral patterns.
• Findings emphasize the significant influence of temporal and infrastructural factors, guiding targeted safety interventions and evidence-based strategies in Jordan.