Investigation of Traffic Accidents Prediction Models and Effective Human Factors: A Review

In most countries of the world, saving human lives is one of the most important and first factors that have been considered by politicians. Among the causes of death, road accident is known as one of the 10 causes of death and casualties. Therefore, paying attention to reducing the number of accidents and also reducing the severity of accidents, is the goal of the country's officials, and their planning and prioritization is in the same direction. One of the most important parts of traffic accident analysis and prediction is selecting and using appropriate models. In this study, extensive research has been tried to give a good view to researchers to choose a suitable model. Also, the human factor in accidents has been studied and the parameters affecting this factor have been studied and the most important ones have been stated. Therefore, the purpose of this study is to examine in detail the most important factors (appropriate model and appropriate parameter) in the evaluation of accidents. Results are shown that Deep learning approach/Data mining/machine learning models had the highest power with 87.27%, followed by Poisson-lognormal and generalized additive models. It was also concluded that most models were used in suburban accidents, however, there were one model “microscopic simulations” that were used more in urban accidents. Deep learning approach / Data mining / machine learning has allocated the most up-to-date research with an average close to 2016 (2015.82). Random-parameters models are next with an average of 13.4. Duration models with the lowest mean (1998.2) are at the bottom of this classification and have the oldest research. Based on this information, it can be concluded that today researchers are more inclined to new models such as Deep Learning, which may be due to the high accuracy of these models.