To address the challenge of feature extraction methods in rolling bearing fault diagnosis relying excessively on human expertise, resulting in suboptimal diagnostic rates, this research introduces a fault detection method that integrates Convolutional Neural Network (CNN) and Support Vector Machine enhanced by the Improved Whale Optimization Algorithm (IWOA-SVM). Firstly, continuous wavelet transform (CWT) is applied to the time-domain signals to obtain time-frequency representations. These representations are then input into CNN for feature extraction. Testing the model on motor drive-end bearing data reveals that the average fault recognition rate exceeds 98.4%. This confirms the practicality and dependability of the suggested approach for diagnosing faults in rolling bearings.