Evaluating road safety improvements becomes important because it can assist policymakers in allocating economic resources to improve safety and implementing effective policy interventions. As such, this study aims to estimate the value of road safety risk measures using a new modeling approach for willingness-to-pay (WTP). Specifically, this study integrates a machine learning technique (decision tree) with a correlated random parameters Tobit with heterogeneity-in-means model. The decision tree identifies a priori relationships for higher-order interactions, while the model captures unobserved heterogeneity and the correlation between random parameters. The proposed modeling framework examines the determinants of public WTP for improving road safety using a sample of car drivers from Peshawar, Pakistan. WTP for fatal and severe injury risk reductions is estimated and used to calculate the values of corresponding risk reductions, which can be used for monetizing the cost of road traffic crashes in the country. Modeling results reveal that most respondents are willing to contribute to road safety improvement policies. Further, the model also uncovers significant heterogeneity in WTP corresponding to the safer perception of the overall road infrastructure and perceived risk of accident involvement. Systematic preference heterogeneity is also found in the model by including higher-order interactions, providing additional insights into the complex relationship of WTP with its determinants. Further, the marginal effects of explanatory variables indicate different sensitivities toward WTP, which can help to quantify the impacts of these variables on both the probability and magnitude of WTP. Overall, the proposed modeling framework has a twofold contribution. First, the modeling framework provides valuable insights into the determinants of public WTP, mainly when the heterogeneous effects of variables are interactive. Second, its implementation and consequent findings shall help prioritize different road safety policies/projects by better understanding public sensitivity to WTP.@en