Predicting the Severity of the effect of car accident on traffic in SFO

This is the final project for our Machine class in collaboration with Alejandro White, Josef Monje, MatMat Romero, Jasper Pangan and Cedric Corro. Findings were shared in an online public presentation.

Background

The city of San Francisco, one of the technologically advanced cities in the US, evidenced by the existence of Silicon Valley, despite having the best economy in the United States is still plagued by the problem of traffic congestion. What is quite alarming is that majority of these tra c congestions are caused by accidents. Around 795 roads in the metro were a ected by accidents. This presents a very drastic problem for the city.

With the advent of technology, plenty of methods can be employed to address the problem regarding trffic, and one such way is the use of machine learning. Assuming traffic is detected and it was found that a recent accident has caused it, some supervised learning methods can help determine if the traffic will be moderate or heavy so drivers can plan their routes ahead of time. This study aims to predict the severity of traffic congestion that is caused by accidents in the city of San Francisco.

About the Data

The dataset used for this study has been extracted from \A countrywide Traffic Accident Dataset” collected by Sobhan Moosavi. The data set includes a multitude of features, however, we decided to reduce it and use only those features which will be available coinciding the car accident.

Insights

The results of this study shows feasibility of using machine learning, specifically tree-based algorithms, to determine the severity of traffic caused by accidents, which is further evidenced by the use of Light GBM which provides a fast, lightweight, and more accurate solution to this problem. The results of the study also provided additional insights to the factors that contribute to the severity of accident-caused tra c congestion. We show in the analysis that some locational attributes and weather attributes such us pressure, humidity, and wind speed a ect how severe congestion will be. These factors are worth looking at to assuage the problem of tra c congestion, much more if it is caused by accidents.