Predicting Severity of US Traffic Accidents: A Machine Learning Approach

Predicting accident severity using data analysis and machine learning is vital for enhancing road safety, as this data can help policymakers and law enforcement gain insights into factors leading to accidents and unsafe driving. Related works in this space have started utilizing machine learning algorithms to build accident severity prediction models. However, despite these models being available, from our insights, we notice that accidents in the United States have been on the rise since 2016, while average accident severity has been on a decline. We propose building an accident severity model using Random Forest and Decision Tree, and we will also perform an exploratory analysis of the data to gain unique insights that may be able to help reduce this rise in accidents. Our findings reveal which States, Cities, and time zones in the United States have the highest occurrences of accidents. Based on this, we recommend that road safety policymakers implement stricter isolated road safety rules to reduce the number of accidents. We also identify how weather conditions, road features, time, and seasonality contribute to this rise in accidents. For example, we notice that most accidents happen between 7 to 8 AM and 4 to 5 PM, so road safety officials can implement stricter road safety rules and better manage their resources during these hours to possibly reduce accidents. The random forest predictive model outperforms the decision tree predictive model, and with our models and insights gained, we hope this information will be of use to policymakers as they try to implement policies to make roads safer for everyone and better manage their resources to reduce the number of road accident fatalities.