Abstract:
The number of vehicles and road transportation increases rapidly daily. Hence the
frequency of road accidents and crashes also gradually increase with it. Analyzing traffic
accidents is one of the essential concerns in the world. Due to the considerable number of
casualties and fatalities caused by those accidents, taking necessary actions to reduce road
accidents is a vital public safety concern and challenge worldwide. Various statistical methods
and techniques are used to address this issue. Hence, those statistical implementations are used
for multiple applications, such as extracting cause and effect to predict real-time accidents. In this
study, a United States (US) Countrywide car accidents data set consisting of about 1.5 million
accident records with other relevant 45 measurements related to the US Countrywide Traffic
Accidents were used. This work aims to develop classification models that predict the likelihood
of an accident is severe. In addition, this study also consists of descriptive analysis to recognize
the key features affecting the accident severity. Supervised machine learning methods such as
Decision tree, K-nearest neighbour, and Random forest were used to create classification models.
The predictive model results show that the Random Forest model performs with an accuracy of
83.95% for the train set and 80.69% for the test set, proving that the Random forest model
performs better in accurately detecting the most relevant factors describing a road accident
severity.
