Deformation caused by clamping in a turning process significantly affects the roundness of the workpiece after machining. Understanding the extent of this deformation beforehand is essential for achieving high-precision machining. However, accurately evaluating the deformation using simple theoretical formulas is challenging. One approach involves predicting the amount of workpiece deformation using 3D models and finite element analysis (FEA). Although this approach can accurately determine workpiece deformation, it requires specialized knowledge and software, which makes it difficult to apply it to a shop setting. To address these issues, this paper presents a method to predict deformation by machine learning the analysis results. The proposed model uses Gaussian process regression with a kernel function designed by considering the physical characteristics of the deformation. The proposed model enhances prediction accuracy by augmenting the training dataset based on prediction confidence intervals. In a validation experiment, FEA was used to determine the roundness of workpieces with randomly selected shapes and materials. A predictive model was created using this as training data. The evaluation confirmed that a properly designed kernel enables high-accuracy predictions even with a small amount of data.