Over the past three decades, more than 8000 pedestrians have been killed in Australia due to vehicular crashes. There is a general assumption that pedestrians are often the most vulnerable to crashes. Sustainable transportation goals are at odds with the high risk of pedestrian fatalities and injuries in car crashes. It is imperative that the reasons for pedestrian injuries be identified if we are to improve the safety of this group of road users who are particularly susceptible. These results were obtained mostly through the use of well-established statistical approaches. A lack of flexibility in managing outliers, incomplete, or inconsistent data, as well as rigid pre-assumptions, have been criticized in these models. This study employed three well-known machine learning models to predict road-crash-related pedestrian fatalities (RCPF). These models included support vector machines (SVM), ensemble decision trees (EDT), and k-nearest neighbors (KNN). These models were hybridized with a Bayesian optimization (BO) algorithm to find the optimum values of their hyperparameters, which are extremely important to accurately predict the RCPF. The findings of this study show that all the three models� performance was improved using the BO. The KNN model had the highest improvement in accuracy (+11) after the BO was applied to it. However, the ultimate accuracy of the SVM and EDT models was higher than that of the KNN model. This study establishes the framework for employing optimized machine learning techniques to reduce pedestrian fatalities in traffic accidents. © 2022 by the authors.