NRGsurf

Description

Using datasets of car accidents and weather in California, prediction factors for car crashes were analyzed. The linear regression predicted the number of accidents per day in Los Angeles to 1.6% using weather data. Random Forest and XGB models yielded similar results of about 90% precision/recall in predicting the factors that determined the severity of car accidents in California.

GOALS

Developing predictive models to identify factors which have an impact on the frequency and severity of car collisions. These models can (i) provide information that can save human lives, (ii) lower insurance costs and (iii) enhance the safety of autonomous vehicles when incorporated into their systems.

DATA SOURCES

Countrywide car accident dataset was sourced via Kaggle. (To access the data a registration on Kaggle is required). The accident data is collected from February 2016 to Dec 2021, using multiple APIs that provide streaming traffic incident (or event) data. For the prediction models and regression, a subset of data for California and Los Angeles were used to avoid computational challenges. The dataset was also reduced to the timeframe from 2016 to 2019 to exclude possible COVID-19 effects.

To avoid endogeneity of the sample, the accident data was enriched with weather data from the LA weather station to include the days when no accidents were reported. The data from this weather station includes a rich set of weather parameters. To access the weather data an API request to the NCEI was generated.