<p>Partitioning discontinuities into subsets is a crucial foundation for assessing the stability of rock masses in engineering. However, the conventional fuzzy C-means clustering algorithm (FCM) is sensitive to outliers and initial cluster centers, requires a predetermined cluster number, and tends to converge to local optima. To overcome these limitations, this paper introduces an optimized fuzzy C-means clustering algorithm based on the density peak clustering algorithm (DPC-FCM) for partitioning rock mass discontinuity sets. The DPC-FCM algorithm leverages global density peak points as potential initial cluster centers, which not only indicates cluster number but also eliminates the randomness in selecting initial centers. This process keeps the FCM algorithm from falling into local optima, achieving global optimization and obviously enhancing the overall robustness of the method. Additionally, by treating points with low local density and far from other high-density points as outliers, the algorithm reduces the impact of outliers and enhances clustering accuracy. The DPC-FCM method was validated using an artificial data set and the San Manual copper mine data set. Comparison results of four widely-accepted methods and three commonly-used clustering validity indexes verified that the DPC-FCM method demonstrated superior rationality and clustering accuracy. Finally, the DPC-FCM method was applied to the separation of discontinuity sets and stability assessments of a deformed rock slope in Mindu Township, China. Combined with the analysis of the geological origin and regional tectonic development history, the method demonstrated its practical effectiveness. The DPC-FCM method is feasible and efficient, enhancing the accuracy and robustness of grouping. It provides a dependable foundation for the 3D discontinuity network model and stability assessment in rock engineering, demonstrating potential for engineering applications.</p>