Abstract: To address interval-valued fluctuations in machine attributes caused by disturbances in uncertain job shops, and the inadequate identification accuracy of conventional algorithms under non-uniform density distributions, we proposes a multi-bottleneck identification method that integrates weighted interval kernel principal component analysis (IKPCA) with mutual nearest-neighbor density peak clustering (MNN-DPC). First, a weighted interval midpoint–radius model is developed, in which a tuning factor is introduced to balance the contributions of machine performance level and variability. Kernel principal component analysis is then employed to extract nonlinear features from interval data and to compute a comprehensive bottleneck index for each machine. Next, an interval distance metric incorporating principal-component information is constructed and fed into an improved MNN-DPC algorithm, where local density is redefined using mutual nearest-neighbor relationships and a two-stage assignment strategy is designed. Finally, the resulting clusters are used to partition machines into hierarchical bottleneck groups. A real-world workshop case study demonstrates that the proposed method not only identifies key bottlenecks consistently recognized by multiple approaches, but also uncovers potential bottleneck machines that are easily overlooked. Moreover, as the disturbance intensity of machine failure rates increases from 2% to 8%, the membership of the Level-I bottleneck cluster remains highly stable, indicating strong applicability to uncertain shop-floor production environments.