A software-based approach to achieve high performance within a power budget often involves dynamic voltage and frequency scaling (DVFS). Consequently, accurately predicting the power consumption of an application at different DVFS levels (or more generally, different processor configurations) is paramount for the energy-efficient functioning of a high-performance computing (HPC) system. The increasing prevalence of graphics processing units (GPUs) in HPC systems presents new multi-dimensional challenges in power management, and machine learning presents an unique opportunity to improve the software-based power management of these HPC systems. As such, we explore the problem of predicting the power consumption of a GPU at different DVFS states via machine learning. Specifically, we perform statistically rigorous experiments to quantify eight machine-learning techniques (i.e., ZeroR, simple linear regression (SLR), KNN, bagging, random forest, sequential minimal optimization regression (SMOreg), decision tree, and neural networks) to predict GPU power consumption at different frequencies. Based on these results, we propose a hybrid ensemble technique that incorporates SMOreg, SLR, and decision tree, which, in turn, reduces the mean absolute error (MAE) to 3.5%.