Aiming at the feature extraction and fault diagnosis of motor rolling bearings, a feature extraction technique based on the starfish optimization algorithm (SFOA) optimized variational mode decomposition (VMD) combined with composite multiscale bubble entropy (CMBE) was proposed. Meanwhile, SFOA was introduced to optimize the kernel extreme learning machine (KELM) fault diagnosis model. Firstly, the parameters of VMD were optimized by employing SFOA, the vibration signal was effectively decomposed into multiple intrinsic mode functions (IMFs). By calculating the Pearson correlation coefficient between each IMF and the original signal, the two components with the largest correlation coefficients were selected. Secondly, the two components were reconstructed and their composite multiscale bubble entropy was calculated to form a feature vector matrix. Finally, the feature vector matrix was input into the SFOA-KELM fault diagnosis model for diagnosis. The experimental results show that the accuracy of this method for fault diagnosis is as high as 100%, and it performs excellently in extracting fault features compared with other models, significantly improving the accuracy of fault diagnosis and having important application value.