Predicting energy consumption accurately and reliably is critical for route optimization in eco-routing. State-of-the-practice methods for calculating energy consumption utilize second-by-second speed, acceleration, and power demand. Such models can achieve high accuracy but are not suitable for forecasting usages due to strict requirement of inputs and computing resources. Other methods used to predict energy consumption rely on average speed data to reduce data collection and computation efforts. However, they ignore the individuality of driving behavior, which is particularly important in near-term predictions of energy consumption, as shown in this paper. This study develops an input-output hidden Markov model (IOHMM) to cope with the influence of external environment and driving behaviors on individual driving features. The model is built and trained using passively collected geospatial location data. The approach furthermore improves the prediction of vehicle specific power (VSP) distribution, a critical parameter for energy predication, through predicted driving features. The model is tested in the Washington D.C. metropolitan area, and the performance is evaluated by comparing various indicators with the real-world values obtained from in-vehicle fuel recording devices. In general, the IOHMM behavior model demonstrates an overall cruising speed accuracy of 86.85% and acceleration rate accuracy of 82.73%. The behavior-integrated energy prediction model outperforms the traditional approaches by increasing the energy prediction accuracy to 86.81%. Results obtained from this study corroborate the importance of behavioral richness, environmental dynamics, and computation efficiency.