A comparison between Artificial Neural Network and Hybrid Intelligent Genetic Algorithm in predicting the severity of...

Area of Science:

• Road safety
• Traffic accident analysis
• Geriatric driving safety

Background:

• Run-off-road (ROR) crashes, particularly those involving fixed object collisions, are a major cause of serious injury and fatalities.
• Elderly drivers (65+) exhibit distinct physiological and psychological characteristics that may increase their vulnerability in ROR crashes compared to younger drivers.

Purpose of the Study:

• To investigate the severity of fixed-object ROR crashes involving elderly drivers.
• To identify key factors influencing crash severity in this demographic.
• To apply and compare Artificial Intelligence (AI) techniques for predicting crash severity.

Main Methods:

• Utilized crash data from California in 2012 from the Highway Safety Information System (HSIS).
• Applied two AI techniques: Artificial Neural Network (ANN) and a hybrid Intelligent Genetic Algorithm (GA).
• Analyzed the predictive performance and feature importance of each AI model.

Main Results:

• The developed ANN model demonstrated superior overall performance.
• The hybrid GA-ANN model showed greater capability in predicting high-severity crashes due to GA's training advantage.
• Light condition was the most significant factor, followed by the presence of right/left shoulders, cause of collision, AADT, number of vehicles, age, road surface, and gender.

Conclusions:

• AI models can effectively analyze ROR crash severity factors for elderly drivers.
• Understanding the influence of light conditions and road geometry is crucial for mitigating elderly driver ROR crash severity.
• Findings can inform targeted safety interventions to improve highway safety for older drivers.