Analyzing, Interpreting, and Forecasting Boston Accident Patterns for a Safer City

Abstract

Using data collected by the city of Boston, this thesis aimed to understand current trends in accident patterns and interpret how the city's current infrastructure contributes to these occurrences. Beginning with an exploratory data analysis of accident patterns in the city, the most dangerous locations in the city were pin-pointed. From there, traffic/congestion patterns were used to further investigate causes of the accidents, or contributing factors. This was also corroborated with time series analyses that teased out time-related patterns in the data and forecast future accident rates. Finally, the effectiveness of current policies were evaluated by measuring their ability to reduce the number of accidents. The results from this analysis confirmed the seasonality of accident patterns and identified the fall to be the most dangerous season, Friday the most dangerous day, and rush hours to be the most dangerous times of day. Upticks in accidents were also observed during meal times, even if congestion and traffic movement around those times were reduced. When removing noise and constructing time series analyses around the data, forecasting models predicted that accident rates will decline slightly over time (assuming there are no major interventions or changes in the city). Finally, current initiatives the city has taken, while not statistically significant, have proven to help reduce accident rates.