Diagnosing errors in roller bearings is an essential part to maintain the normal operation of modern machinery, especially under different operating conditions. This research introduces a new method for roller bearing fault diagnosis based on the Support Vector Machine (SVM) with parameters optimized using the Hunger Games Search (HGS) algorithm, referred to as HGS-SVM. Firstly, acceleration vibration signals were decomposed into component functions by empirical Fourier decomposition (EFD) method. Secondly, these functions extract initial feature matrices using the singular value decomposition (SVD) method. Finally, these values were used as initial vectors for the HGS-SVM classifiers. The results of this paper show that the proposed method has a lower test error rate and short time cost when compared with other methods with the same collected data.