Application of extreme value theory models to crossing conflicts in mixed traffic intersection. Evidence from a case study in Yaoundé, Cameroon

The limitations of the crash-based safety approach have led to the evolution toward more innovative and proactive approaches such as traffic conflicts, which are non-crash events described using surrogate safety measures, used to estimate crashes through different statistical models. Extreme Value Theory (EVT) models have been applied for over two decades to study the relationship between traffic conflicts and crashes and have seen significant and promising applications in High-Income Country (HIC) contexts. Low- and Middle-Income Countries (LMICs), characterized by poorer crash data quality, could benefit from this proactive technique. However, there is still limited evidence of the application of conflict-based safety analysis in general and EVT models in LMICs. The difference in application between HICs and LMICs is related to the unique specificity of LMICs, posed by their less organized, non-lane-based, heterogeneous traffic conditions. This paper seeks to bring new evidence of the application of EVT in an LMIC, Africa context. Data were collected with video to observe crossing conflicts at road intersections between motorized and vulnerable road users. The video was processed to extract conflicts using Post-Encroachment Time (PET) and speeds. Two EVT models were used, namely the Block Maxima approach and the Peak Over Threshold approach. Mean residual plots were used to obtain threshold values, and the model’s parameters were obtained with maximum likelihood estimation. Preliminary results suggested that the Peak Over Threshold models tend to perform better than Block Maxima models. These results show the potential of EVT models to estimate crashes in mixed traffic.