Abstract: As a widely used alloy structural steel in gear and shaft components, the hardenability of 8620 steel significantly influences the hardness distribution and mechanical properties of the final product. Traditional methods for assessing hardenability rely heavily on extensive experimental testing, which is both time-consuming and costly, and fails to meet the demands of modern industry for efficient and accurate predictions. In this study, 834 hardenability data points were preprocessed by analyzing the chemical composition and Jominy end-quenching hardness data. The SHAP method was employed to evaluate the importance of individual features, revealing the significant influence of elements such as Cr, C, Mn, Mo, Al, and N on hardenability. Following feature screening, a machine learning prediction model for the hardenability of 8620 steel was developed to achieve high-precision prediction of the J7.9 value. This approach provides a scientific basis for alloy design and optimization of heat treatment processes in steel manufacturing.