Abstract: In response to the challenges posed by imbalanced samples leading to low recognition accuracy and high feature redundancy in AC contactor, a novel composite recognition methodology which is leverages embedded random forest (ERF) and Bayesian optimization-support vector classification (BO-SVC) is introduced. Firstly, the extraction of contactor state features from the full life testing platform designed for contactor is initiated. To counteract the low recognition accuracy caused by the imbalance among different state samples, a sample balancing strategy based on the weighted method is proposed. Subsequently, the ERF is employed to perform feature selection and reduction on the balanced samples. This process leads to the extraction of optimal features that represent the dynamic patterns of AC contactor state changes. Following the feature extraction step, the selected optimal features are fed into BO-SVC recognition model. A comprehensive evaluation of BO-SVC's fault recognition capabilities is undertaken, compared with two other representative models, the performance of each model is evaluated based on three indicators: accuracy, recall, and F1-score. The results of the proposed method reaches 95.22%, 98.91%, and 97.01%, respectively, all of which are higher than the comparison models. Using F1-score as an indicator, the performance of each model is tested on four sets of samples, and the results showed that the F1-score of the proposed method is on average 0.56% and 27.28% higher than the compared models, respectively. The research in the article effectively solves the problems of redundant characteristics and low fault recognition accuracy of AC contactors.