This project explores time series forecasting of daily traffic crash rates in Chicago from 2018 to 2024, with a focus on understanding how past crash patterns and external conditions influence future risk. The primary research question asks: To what extent does yesterday’s crash rate help predict today’s? Using a combination of Holt-Winters exponential smoothing, Prophet forecasting, and SARIMAX models, we assess the role of autoregression, seasonality, and exogenous variables such as weather and roadway conditions. Daily crash data was cleaned, aggregated, and enriched with engineered features including holiday indicators, weather metrics from O’Hare and Midway airports, and binary flags for poor lighting, lack of traffic controls, and road defects. Modeling results show that crash rates exhibit strong weekly seasonality and moderate predictability from lagged values. While external variables improved forecast accuracy marginally, the most reliable signal came from temporal patterns themselves. These findings suggest that structured, short-term forecasting of traffic crashes is feasible and could support proactive safety measures in urban planning and emergency response.
↧