2nd Place Winner in Fall 2018
by Wen Ying, Teng Li and Jinyu Chen
This document is about Crash Type and Incidence of Accidents Prediction in Vehicle Crashes, which is a project of CS 6316 - Machine Learning. Vehicles crashes is one of the most important problems because of the increasing number of cars and drivers. And many people died or got injured in these crashes, and they also lead to great economic losses. So it is meaningful that we should dig more information from the crash data and find some rules to control or even improve the traffic. Actually, machine learning, as a popular method of processing data, can certainly help the problem of car crashes. In this document, we firstly implement data visualization by plotting and Quantum Geographic Information System (QGIS)1, which can help us find any correlation of information. Then we apply different classification methods on crash data after data preprocessing, including SVM, Boosting and Random Forest. Eventually, we obtain a best accuracy of 67.0% to distinguish 16 categories of crashes. Thus to decide the type of car accidents automatically, aim at helping the police to rescue people and deal with crashes more efficiently. This can also give a warning to government and drivers to focus on possible car crashes in the future. In addition, this document differentiate the crash data into 5 clusters by K- Means, which can alert people in advance that certain places have high incidence of accidents..
Comments