<p>In the field of vibration monitoring for rotating machinery, high-precision bearing fault diagnosis faces challenges due to imbalanced measurement data and noise interference from complex environments. To address these issues, this paper proposes a bearing fault diagnosis method based on an ensemble undersampling extended belief rule base (EBRB-EU), aiming to improve the reliability and accuracy of equipment condition assessment. Specifically, to overcome the limitation of using Euclidean distance in EBRB, which fails to effectively capture the differences in the probability distribution of vibration signals, the paper introduces the Kullback-Leibler (KL) divergence to replace the Euclidean distance, providing a more accurate measure of the match between samples and rule antecedent attributes. An ensemble undersampling strategy is then employed to calibrate the training set, reducing the impact of majority class samples on model training. Subsequently, the data is divided into multiple subsets, and a set of parallel sub-EBRB models is constructed. The differential evolution algorithm is used to optimize the parameters of each sub-model, minimizing evaluation bias. Finally, an ensemble voting mechanism is applied to integrate the diagnostic results, enhancing the accuracy of fault state identification. Experimental results demonstrate that the KL divergence-based EBRB-EU model outperforms the Euclidean distance-based EBRB method on multiple datasets, effectively alleviates the data imbalance issue, and exhibits strong resistance to interference.</p>