Low-voltage switch fault diagnosis is challenged by imbalanced fault samples and insufficient recognition of minority classes. This study proposes a hybrid SVM-SMOTE-AdaBoost-RBF (SSAR) model that combines SVM-guided minority-class oversampling with adaptive ensemble-based nonlinear classification. In the proposed framework, SVM-SMOTE uses decision-boundary information to support minority-class sample balancing, while AdaBoost-RBF adaptively adjusts sample weights and performs fault classification. Experimental results show that SVM-SMOTE achieved a data completeness rate of 96.45% on the CWRU-CB dataset and a fault-type recognition accuracy of 98.11% on the NASA-Relay dataset. On the self-generated dataset, SSAR achieved a diagnostic accuracy of 98.24%, a precision of 97.11%, a recall of 95.98%, and an F1-score of 96.48%, with a diagnosis time of 2.08 s for 100 iterations. These findings indicate that integrating boundary-guided sample balancing with adaptive nonlinear classification can improve fault diagnosis under imbalanced data conditions. Further validation under realistic operating conditions is required to establish its practical applicability.