Despite the availability of energy expenditure (EE) information from smartwatches, estimation error rates remain high, particularly when measuring a wide range of running speeds. To overcome this inaccuracy, we proposed a method based on biomechanical characteristics relation between mechanical work and EE. The main process consists of two major steps. First, using a single inertial measuring unit (IMU) and machine learning, we estimate ground reaction force and derive three biomechanical components (contact time, propulsion force, and mechanical power) that are known to be strongly related to EE. Second, we estimate EE at various running speeds using the derived biomechanical components. As a result, we could estimate EE under 20% MAE in varying ranges of running speed. In conclusion, this study demonstrates the potential of using biomechanical knowledge to enhance the quality of provided information while keeping wearable devices practical.