In this project, we implemented several classical machine learning algorithms to predict the severity of car accidents in the US. We also used LSTM, a special recurrent neural net, to analyze the accident description and predict the accident severity. We achieved a best accuracy of 92%.