Statistics and Modeling of the Severity of Motorcycle Accidents on The Roads of Gilan Province Through Regression and Artificial Neural Network Method

Accidents impose irreparable costs on the economy of countries every year. In addition, many of the death statistics are related to accidents, which can endanger the human factor itself, which is the driving factor of the economy. In the meantime, one of the road users who is very are exposed to accidents, it is the motorcycle riders who every year include a large part of the dead and injured in the accidents, the main purpose of which is to investigate the variables affecting the accidents of motorcycle riders. Collecting information on accidents as well as using various statistical and modeling methods such as Friedman, Aamili, Logit model and neural network to investigate the effect of each of these variables. The results showed that the variables of accident season, cause of accident, road surface conditions and lighting conditions had the highest rank respectively. On the other hand, the variables of the day and season of the accident and the age of motorcyclists, respectively, had the least importance in the occurrence of motorcyclist accidents on the roads of Gilan province. In addition to these variables, weather conditions, road surface conditions and lighting conditions were the first effective factors in the accidents of motorcyclists in Gilan province; According to the logit model obtained from the accidents of motorcycle riders, it was found that the variables of spring season, 6 to 12 and 18 to 24, dry surface conditions, night lighting conditions without sufficient light; rainy weather; Failure to pay attention to the front and sudden change of direction increase the possibility of accidents of motorcyclists on Gilan road crossings. based on the neural network model; The values of cause, lighting conditions and weather conditions respectively have the greatest effect on the severity of motorcyclist accidents on Gilan roads. Also, the neural network model with R2 values of 93.31 showed higher accuracy than the logit model.