DC2. Explainable AI for road safety: benchmarking AI methods and data

Hosts: TUD & AGILYSIS

Objectives:

• To define the needs for explainability in AI for road safety, from the perspective of policy makers (transport authorities)
• To disentangle strengths and weaknesses (prediction accuracy and interpretability) of two types of AI methodologies –ML algorithms
• and statistical/econometric models
• To understand the performance and optimally integrate both techniques with benchmark datasets and applications in road safety for AI
• transparency in decision support

Expected results:

• A taxonomy of explainable AI methodologies and their applications in road safety
• A model-agnostic methodological framework for mixing ML algorithms and statistical/econometric methods
• A new explainable AI-based risk mapping tool for urban roads in West Midlands, UK

Planned secondment(s):

• Transport for West Midlands, Purpose: to incorporate the decision maker’s perspective on explainable AI and collect data for testing the explainable AI framework.