الفهرس 
Title : Severity prediction of traffic accident using machine learning models
ContentOnly 14 pages are availabe for public view
 |  | from | 62 |  |  |
| | | from | 62 | | |
Abstract
Car accidents, or what is also known as traffic accidents, occur as a result of a collision between a vehicle and another vehicle, and traffic accidents cause severe damage and serious injuries, some of which may be minor so that their results are only material damage, and sometimes they may be serious. Each year, approximately 1.25 million people lose their lives as a result of traffic. Between 20 million to 50 million people suffer non-fatal injuries and many become disabled as a result. Machine learning is one of the branches of computer science, which aims to give the computer the ability to learn and come up with conclusions and results without the need to be programmed to do so. This study firstly aimed to explore the performance of several machine learning models in predicting the severity of road accidents. Second, the comparison between the results of machine learning models and statistical models. Third, identify the main factors that contribute to the severity of accidents. Finally, predict the severity of the injury for any new accident. The study found that fatal and severe accidents are a relatively small proportion of the total accidents 16.67% and 14.72%, respectively. The age group between 18 and 35 years is the most prone to traffic accidents. Accidents most commonly occur on weekends (Thursday, Friday and Saturday). Regarding the speed of the car, the highest fatalities occur at speeds of 110, 120 and 140 km / h. It was found that users of bicycles are less likely to be injured and die. Sudden deviation, not leaving enough distance, not adhering to the mandatory lane, crossing a red light and speeding without taking into account road conditions are the most common causes that lead to road accidents. Several machine models were used such as Support Vector Machine, Naïve Bayes, XGBoost, Random Forest, Decision Trees and the Multinomial Logistic Regression statistical model, showing the accuracy of the models performance on the data 55%, 64%, 71%, 72%, 49% and 61% respectively. The findings revealed that Random Forest classifier outperformed other machine learning algorithms and statistical models.