Daily Multi-Level Prediction of Traffic Accidents Using Hierarchical Time Series

Traffic accidents are a cause of deaths in Brazil and worldwide. This impacts road transportation, which is fundamental for the transport and economy of various countries. This study analyzes the prediction of traffic incidents on Brazilian federal highways, using traffic incident response data between 2017 and 2023, aiming to reduce these incidents. A daily predictive approach was developed with strategic (365 days), tactical (30 days), and operational (7 days) perspectives. Organizing the data in time series, nine optimization algorithms were used and compared. The methodology included reconciliation techniques to adjust forecasts between different hierarchical levels. The results demonstrated high predictive capacity, with a minimum long-term explainability of 84.5\%, reaching 99,\% in short-term forecasts. The model showed potential applicability for road safety agencies, allowing efficient predictive actions to reduce this social problem.