Abstract The emergence of new transportation modes, also new challenges for modeling traffic system, have been the motivation for the development of more robust traffic simulation models. An overall framework for estimation of data-driven models is developed. Computational approaches, such as clustering, classification, and regression techniques, are employed in a novel way and model components are analyzed. The data-driven approach is first validated on two conceptual case studies: one for mesoscopic modeling and another one for microscopic modeling. Application to weak lane discipline modeling is also attempted and the feasibility is evaluated. Finally, a network-wide application is presented using the microscopic simulator SUMO and implementation aspects are discussed. Data-driven estimation of traffic simulation models appears to be a promising tool appropriate to adapt to new transportation challenges.