A well-designed energy management system plays a crucial role in increasing fuel efficiency and reducing polluting emissions in dual-power hybrid electric vehicles (HEVs), which are an intermediate stage in the transition from combustion engine vehicles to fully electric vehicles. Despite many studies to optimize energy management, innovative ideas are needed to ensure the most appropriate energy use according to changing road, vehicle, and driver types. For this purpose, we developed a data-driven method to construct a stochastic energy management system, considering realistic uncertainties. We have demonstrated that an HEV can be used more efficiently with an appropriate energy management strategy depending on the road type and driving style. We collected and analyzed 38 thousand km of real driving data with nine different drivers. We transformed these data into meaningful information with a comprehensive data processing methodology and then classified driving styles according to these data using data mining methods. The classification algorithm we designed predicted driving style for three different roads with an average success rate of 95%. We achieved better fuel and emission values with a fuzzy logic-based energy management system that we designed according to the driving style determined by our classification algorithm. The fuzzy controller we developed achieved fuel improvements of up to 7% on the motorway, 9% on the urban road, and 16% on the residential district, based on real driving data results. Although there is a trade-off between fuel and pollutant emissions, our proposed system has also produced significant improvements in harmful emissions. Our results can be used as an inspiration and guide in the studies of improving fuel and emissions in HEVs.