Regional Classification of Dominant Discontinuity Sets in Open-Pit Rock Slopes: A Case Study of the Dabaoshan Mine

Regional classification of dominant discontinuity sets supports open-pit slope characterization, but dip direction is circular and should not be analyzed as a linear variable. We reanalyzed 73 area-level records from 72 nominal survey areas at the Dabaoshan Open-Pit Mine; Area 9 contained two subareas. Dip direction was encoded by jointly scaled cosine and sine components, and three linear descriptors were standardized. Candidate solutions for K = 2–10 were evaluated using internal validity indices, minimum cluster size, and 300 bootstrap refits. The gap one standard error rule favored K = 2. K = 5 was retained as an exploratory engineering resolution because it had the best silhouette coefficient (0.283) and lowest Davies–Bouldin index (1.154) among K = 2–7 while retaining at least five records per cluster. Its Calinski–Harabasz index was 21.844, and bootstrap stability was moderate (adjusted Rand index, ARI = 0.643 ± 0.168). Circular mean directions ranged from 80.3° to 244.1°, with significant separation in 20,000 permutations (p < 5 × 10−5). A five-unit radial basis function (RBF) network reproduced the K-Means pseudo-labels with 97.7% ± 3.8% repeated cross-validation fidelity. This measures partition reproducibility, not independent predictive accuracy. The workflow supports structural domain screening; site-specific stability decisions require geometric, hydrogeological, and kinematic evidence.