A genetic programming approach to explore the crash severity on multi-lane roads.

Area of Science:

• Road safety
• Traffic engineering
• Machine learning applications

Background:

• Understanding factors contributing to road traffic crashes is crucial for developing effective safety interventions.
• Traditional methods for crash analysis often develop single models, limiting the exploration of complex interactions.

Purpose of the Study:

• To investigate the relationship between geometric and environmental factors and injury-related or severe crashes.
• To develop and apply advanced classification models for crash analysis.

Main Methods:

• Utilized Linear Genetic Programming (LGP), an evolutionary computation technique that evolves computer programs.
• Employed Discipulus software to evolve multiple classification models through biological evolution concepts, including crossover and mutation.
• Compared LGP's multi-model approach with traditional single-model methods like classification and regression trees.

Main Results:

• Vision obstruction identified as a primary factor in severe crashes.
• A higher percentage of trucks, even if small, increases the likelihood of injury-prone crashes.
• Safe median designs ('lawn and curb') are associated with reduced angle/turning crashes.
• Favorable conditions such as dry surfaces, good pavement, wider shoulders, and sidewalks decrease crash severity.
• Interactions, like on-street parking combined with higher speed limits, elevate the probability of crash injuries.

Conclusions:

• Linear Genetic Programming offers a robust method for identifying complex factors influencing crash severity.
• Specific geometric features and environmental conditions can be modified to mitigate severe and injury-related road traffic incidents.
• The study highlights the importance of considering interactions between variables for comprehensive road safety strategies.