Assessing the Transferability of the Highway Safety Manual Predictive Method for Urban Roads in Fortaleza City, Brazil

Assessing the transferability of crash prediction models for two lane highways in Brazil

“…Though, only few researches performed an examination of transferability for safety analysis. Previous studies found out four more significant metrics (Al Kaaf & Abdel-Aty, 2015;Cafiso, Di Silvestro, & Di Guardo, 2012;Cunto et al, 2013;D'Agostino, 2014;Haas, 2015;Sacchi, Persaud, & Bassani, 2012;Waihrich, 2015;Washington, Persaud, Lyon, & Oh, 2005): (1) The R 2 can provide a measure of how well the observed crashes are likely to be predicted by the model. This metric was addressed in most studies due to its popularity and simplicity (Al Kaaf & Abdel-Aty, 2015;Cafiso et al, 2012;Cunto et al, 2013;D'Agostino, 2014;Silva, 2012;Waihrich, 2015).…”

“…There is a gap in the knowledge related to CPM models for Brazilian conditions (Cunto, Sobreira, & Ferreira, 2013;Silva, 2012;Waihrich, 2015) and modeling crashes phenomena is an important step to avoid future accidents and consequently unnecessary deaths. Moreover, an evaluation of the transferability of models is always desirable to improve safety analysis and take advantage of the most recent models.…”

“…For multilane highways of the states of Minas Gerais, Goiás and Distrito Federal due to the small size of the sample used, the calibration of HSM model showed better results for Minas Gerais than for the combination of Goiás and Distrito Federal (Waihrich, 2015). For urban intersections in the city of Fortaleza, the HSM was used for estimating the expected average crash frequency, regarding its transferability, and the study suggested that the HSM should be used with caution (Cunto et al, 2013).…”

“…The mean absolute percentage error (MAPE) has been used for some researches to assess the predictive performance avoiding bias due to different segments with more accidents (Al Kaaf & Abdel-Aty, 2015;Cunto et al, 2013;Waihrich, 2015). Through Equation 2.10 MAPE can be obtained:…”

“…Most studies used Mean Absolute Deviation (MAD) as Goodness-of-fit (GOF) measure (Al Kaaf & Abdel-Aty, 2015;Cunto et al, 2013;Persaud, Saleem, Faisal, & Lyon, 2012;Sacchi et al, 2012;Waihrich, 2015). It differs from MAPE because it can consider segments where no crashes happened, which means observed values were equal to zero.…”