Detecting Cherry-Picking and Quantifying Uncertainty in Road-Accident Analytics for Reliable Public-Safety Decisions
Dr. Patrick Niyishaka holds a PhD in Computer Science from the University of Hyderabad, specializing in computer vision and digital image forensics, and an M.Sc. in Software Engineering. He is a computer and data scientist based at Kepler College in Kigali, Rwanda.
His research focuses on improving how road-accident information is interpreted so that public-safety decisions rely on clear, transparent, and trustworthy evidence. Road-traffic crashes remain a major global health challenge, causing approximately 1.19 million deaths and 50 million injuries each year. They are the leading cause of death among people aged 5 to 29. Africa records the highest road-traffic fatality rates worldwide. In Rwanda alone, about 9,600 accidents and 350 deaths were reported in 2024, directly influencing national safety strategies and infrastructure planning.
Dr. Niyishaka previously analyzed Rwanda’s accident data from 2010 to 2015 through the AutoRTC project. However, accident analytics face two major risks: analytic cherry-picking and weak uncertainty estimation. Cherry-picking occurs when reports selectively highlight certain districts or time periods, potentially misleading policymakers. Weak uncertainty estimation arises when accident figures are treated as exact, even though real-world data may contain reporting gaps or inconsistencies.
As a Schmidt AI in Science African Faculty Fellow, he will develop an AI-powered system that detects cherry-picking and integrates uncertainty estimation into accident trend analysis. Using Bayesian Linear Regression and Bayesian Neural Networks, the system will model seasonal patterns, day-night differences, sudden spikes, and the influence of weather and road-user type.
By integrating AI agents and multilingual language models, the project will generate clear and accessible explanations for stakeholders. This work aims to strengthen Rwanda’s road-safety systems and create a scalable framework for Africa and beyond.