Traffic accidents are one of the most important causes of death and disability in the world, which cause much damage to countries and millions of people every year. Many researchers have focused on the time series caused by different aspects of these accidents. Box-Jenkins is one of the most common and widely used time series forecasting models, which is used under the condition of model stationarity. With the advancement of technology and the development of neural networks in time series, the discussion of comparing the predictive power of these models with traditional time series models has been raised. In this study, we investigated the traffic accidents leading to death or injury in iran during 2011 to 2021.

Materials and methods: Statistical analyses related to the SARIMA time series model have been performed by using Minitab and EViews and the LSTM neural network model by Python and Visual Studio Code.

Results: During period of the study, The trend of the number of deaths has always been decreasing, but the number of injured has been increasing before the coronavirus epidemic and decreasing after that. The highest number of dead and injured has always been in Shahrivar (August and September) every year and the months with 9 days off had significantly less deaths and more injuries. 21.38% and 27.56% of dead and injured were women. The ages of 20 to 29 had the highest number of deaths and injuries, but children and the elderly were more vulnerable and their share in the deaths was more than the injuries. According to the results, as the weather worsens, the share of accidents leading to death increases. In comparison of SARIMA and LSTM the results indicated that LSTM has been better in estimating data trends. The forecast shows the increasing trend of fatality and injuries in the comind years.
Conclusion: Compared to the SARIMA model, the LSTM model showed better performance in predicting trend and time series components.