Abstract: Objectives To address the insufficient accuracy of conventional frequency-domain fatigue analysis methods under wide-band random fatigue loading, a machine learning-based method for frequency-domain calculation of wide-band random fatigue damages is proposed. Methods First, a dataset relating a wide range of spectral moment parameters and spectral width parameters to the corresponding rainflow counting fatigue damage is constructed based on 11 types of parameterized stress spectra for training, hidden layer structure optimization and performance evaluation using a neural network based machine-learning model. Secondly, numerical simulation validation is conducted using independently generated set of new stress spectra to systematically assess the prediction accuracy and generalization ability of the model. Thirdly, SHAP (SHapley Additive exPlanations) analysis is conducted to interpret the internal learning mechanism of the model. Finally, the applicability of the model under extreme spectral width conditions and real engineering scenarios is examined through extreme spectral width cases and actual engineering simulation data. Results The results under all validation cases demonstrate that the proposed method generally outperforms traditional frequency-domain methods in prediction accuracy. In most cases, the maximum relative error is maintained within 5%, with the mean absolute percentage error below 1%; the average computational time per prediction is approximately 4.5 ms, indicating high computational efficiency while ensuring reliable fatigue damage prediction. SHAP-based interpretability analysis reveals that the fourth-order spectral moment serves as the dominant feature driving the model predictions. Conclusions The proposed ANN model effectively captures the complex nonlinear mapping relationship between spectral moment and fatigue damage, overcoming the limitation of traditional frequency-domain methods that are constrained by analytical formulations and cannot adequately characterize complex spectral moment features. Consequently, the proposed method achieves high-efficiency and high-precision prediction of broadband random fatigue damage.