Energy consumption in video decoding is a complex interplay of codec efficiency, hardware design, software optimization, and other factors such as the processor clock frequency for software implementation. The Dynamic voltage and frequency scaling (DVFS) allows adjusting dynamically the processor clock frequency according to estimated frame requirements, enabling efficient resource management for decoding tasks. However, decoding time can vary considerably from one frame to another due to variations in frame complexity. Building upon this concept, this paper proposes a machine-learning model that estimates the decoding time of individual frames. The proposed model is built on the ExtraTrees regressor that accurately predicts the decoding time of 1080p video frames with a low relative error of 5.58% and a high R2 score of 94%. Our proposal entails the utilization of frame-related data that is readily reachable by the decoder, making it highly suitable for real-time scenarios.