Abstract: Circuit breakers for electric vehicles are critical components for electrical safety, which suffer from various defect types with weak early-stage fault features. This method establishes a circuit breaker defect classification model integrating the Moving Average Filter (MAF) and decision tree to realize circuit breaker fault diagnosis. Firstly, the Moving Average Filter is adopted to smooth the circuit breaker current dataset. Secondly, eight key features are extracted from the smoothed dataset as model inputs, and four delay-related parameters are set as outputs to construct the training dataset. Finally, the MAF-decision tree defect classification model is established, and its performance is compared with the SVM and KNN algorithms.Simulation results show that after one hundred random partitioning tests, the average accuracy of SVM, KNN and the original decision tree reaches 80.30%, 86.22% and 92.90%, respectively. After applying the MAF smoothing filter, the average accuracy of MAF-SVM, MAF-KNN and MAF-decision tree rises to 90.17%, 93.60% and 97.95% correspondingly. Experimental results verify that MAF can effectively remove signal noise, and the decision tree achieves higher classification accuracy than SVM and KNN. The MAF-decision tree method possesses outstanding defect classification performance.