<p>To address the challenges of degradation indicator construction and insufficient sensitivity for early fault detection in rolling bearing performance degradation assessment, this paper proposes a novel method for bearing degradation evaluation, which integrates kernel optimal global–local preserving projection (KOGLPP) and Sinkhorn distance. First, a multidimensional feature set is constructed by extracting time-domain, frequency-domain, and entropy features from the original vibration signals of the rolling bearings. Then, a comprehensive selection criterion based on monotonicity, correlation, and trend is developed to select effective degradation features and construct a sensitive feature set. Next, the sensitive degradation indicator KOGSD is constructed by combining the KOGLPP and Sinkhorn distance methods. Finally, the health threshold is determined based on the Chebyshev inequality principle for early fault detection and degradation assessment. This method introduces optimal transport theory into degradation feature modeling and overcomes the limitations of traditional methods in handling vibration signals by combining global structure and local geometric properties, improving sensitivity to nonlinear features. Experimental results from the IMS and XJU-SY datasets, as well as actual engineering cases, demonstrate that the proposed method exhibits higher accuracy and sensitivity in representing bearing degradation trends and detecting early faults, compared to traditional degradation indicators. The results show that the proposed method outperforms traditional methods in terms of early degradation point identification and the number of false alarms.</p>