Forecasting road traffic accidents accurately is crucial for safety planning and policy. ARIMA modeling is frequently used for forecasting road traffic accidents; however, researchers vary on interval year-basic and predicted. This research aims to identify ARIMA’s accuracy in forecasting traffic accidents with variation in year-basic and predicted use of 10-year road traffic accident data in Surakarta, Indonesia. The P-value is used to consider the fitted model, and the Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) are used to assess the accuracy of the model. Of the research results, only four can be modeled for a one-year interval among ten data, with all the P-values for the fitted model being less than 0.05. All data can be modeled for two-, three-, and seven-year intervals, with all P-values less than 0.05. Considering the MAE and MAPE rate during the COVID-19 pandemic, it is found that the MAE and MAPE values are much higher when the prediction includes year-basic and predicted during the COVID-19 pandemic than when the prediction is based on the year-basic and prediction are all in normal conditions.