<p>One factor limiting machining efficiency during milling is the spindle power cap. For high-efficiency machining, predicting spindle power in advance is crucial, enabling the use of cutting parameters that maximize material removal without exceeding this limit. In this study, we developed a machine learning model using Gaussian process regression to predict spindle power from actual machining tests. Collecting extensive data during these tests is difficult due to the required resources, such as materials, tools, and time. Although simpler machine learning models often suffer from reduced accuracy in data-sparse regions, Gaussian process regression maintains high predictive accuracy even with limited data by adapting the kernel function. We selected kernel functions that align with the theoretical formula for spindle power. A linear kernel represented the proportional relationships among the explanatory variables, while an RBF kernel captured non-linear relationships, such as exponential functions. First, we used a simple mathematical model to evaluate the appropriate kernel combination. Next, we trained a predictive model with data from actual machining tests and evaluated the effectiveness of this kernel combination. The resulting model accurately predicts spindle power, even in areas with sparse data or requiring extrapolation.</p>