This study proposes the Kolmogorov-Arnold Network (KAN) as a novel approach for predicting surface roughness (Ra) and cutting force (Fc) in the milling of SKD11 steel, a material widely used in die and mold manufacturing. Unlike conventional machine learning models, KAN yields interpretable symbolic expressions that capture the nonlinear relationships between machining parameters—cutting speed, feed rate, and depth of cut—and the resulting Ra and Fc. Experimental validation on 27 milling tests demonstrates that KAN attains high predictive accuracy, with RMSE values of 0.0075 for Ra and 0.0177 for Fc, outperforming traditional regression methods. Feature importance analysis reveals that feed rate and axial depth of cut have the greatest impact on Ra, while feed rate, axial depth, and cutting speed collectively dominate Fc. These findings highlight KAN’s potential to enhance process optimization, reduce trial-and-error experimentation, and improve manufacturing efficiency in machining operations.