Assessing the transferability of the Highway Safety Manual crash prediction model for divided rural multilane highways in Brazil

• Autor(a) principal: Mendes, Olga Beatriz Barbosa
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/18/18144/tde-19092023-101215/
• topic: Highway Safety Manual; crash prediction model; fator de calibração local; Highway Safety Manual; local calibration fator; modelo de previsão de acidentes; road safety; segurança viária; transferability; transferibilidade

The predictive method presented in the Highway Safety Manual (HSM) estimates the crash frequency by combining a safety performance function (SPF) with crash modification factors (CMFs) and a calibration factor to consider local conditions. This study aims to assess the performance of HSM predictive models when applied to a different condition, such as found on Brazilian roads, by evaluating rural multilane highways. Five divided four-lane highways were segmented and considered following HSM guidelines. To deal with the possible underreporting number of Property Damage Only (PDO) crashes, further investigation was developed for total and Fatal or Injury (FI) severity. Calibration factors (Cx) were calculated, 2.62 for total and 2.35 for FI crashes. The goodness of fit (GOF) tests applied were Mean Absolute Deviation (MAD), Mean Absolute Percentage Error (MAPE), Mean Squared Error (RMSE), R2, and observed versus estimated collisions graphs for different scenarios. The Goodness of Fit measures to assess the HSM performance shows that models for total crashes perform better than FI. Finally, as 2020 was an atypical year in which the COVID-19 pandemic altered traffic patterns worldwide, this study aimed to assess the application of the calibrated prediction model to a sudden disturbance in traffic behavior. The HSM method was applied to 2020 using the Cx obtained from the four previous years. The result showed that for 2020, the observed counts were about 10% lower than the calibrated predictive model estimate of crash frequency for all types of crashes. However, the calibrated prediction of FI crashes was very close to the observed counts.