SILO: Highlights from UW–Madison’s data science partnership with American Family Insurance

Welcome: Ramya Korlakai Vinayak, Assistant Professor, Electrical and Computer Engineering

Talk 1

Title: Safer Driving Through Optimized Telematics-Based Feedback
Presenters: Fengxu Li, undergraduate student, Industrial and Systems Engineering & Rahul Shenoy, PhD student, Industrial and Systems Engineering
Abstract: This study evaluated a behavioral analytics algorithm that delivered personalized, telematics-based nudges to reduce unsafe driving behaviors in a 6-month naturalistic driving study with 18 participants. Using Thompson sampling adapted for non-stationary behavioral change, the algorithm selected from five nudge types, Educational, Social Proof, Curiosity, Standard, and No Nudge. Nudges were associated with a 43.6% reduction in hard braking events (IRR = 0.56, p = 0.001), with 89% of participants showing improvement. Results highlight that personalization is a critical driver of effectiveness, as individual responses to nudge types varied, suggesting broad potential for application in fleet management, insurance, and consumer telematics.

Talk 2

Title: Vision-Language Model for Driving Environment Risk Analysis
Presenter: Yiwei Zhang, PhD student, Computer Sciences
Faculty Advisor: Song Gao, Associate Professor, Geography
Abstract: Accurate and reliable assessment of driving environment risks enhances road safety and supports efficient urban transportation management. This talk will introduce a novel GeoAI framework that employs Vision-Language Models (VLMs) with a knowledge-guided Retrieval-Augmented Generation (RAG) system to better evaluate such risks. 

Talk 3

Title: Simulating Hurricane Migration: A Digital Twin Approach
Presenters: Lauren Khoury – M.S. in Data Science in Human Behavior, Jake Murray – M.S. in Data Science in Human Behavior, M.S. in Data Science in Human Behavior, Ho Wong – M.S. in Data Science in Human Behavior
Abstract: This project applies a digital twin framework to model migration decisions in hurricane-prone areas. We combine demographic, economic, and hurricane risk data to build generative agents informed by machine learning modeling and feature importance analysis. By simulating responses to a Category 5 hurricane scenario, we analyze how financial vulnerability, prior experience, and risk perception may influence migration intent.