Spencer-Method Slope Stability Analyses for Machine Learning Surrogate Training: Seed, Continued LHS, Surrogate-Guided, and Test Sets

This dataset supports the study of surrogate-guided sampling for machine learning prediction of slope stability. It comprises four files of two-dimensional, homogeneous slope configurations analyzed by Spencer's limit-equilibrium method in HYRCAN, covering a five-dimensional parameter domain (slope angle 10–65°, friction angle 5–40°, cohesion 1–75 kPa, unit weight 14–22 kN/m³, slope height 1–25 m). The dataset is the complete output of a budget-fair comparison between Latin hypercube and surrogate-guided sampling strategies, as described in the associated publication. Files seed_lhs_300.csv — Initial Latin hypercube seed of 300 Spencer-method analyses, used to train the preliminary surrogate model and as the shared starting point for all experimental arms. lhs_continued_300.csv — An additional 300 independently drawn Latin hypercube analyses (arm B), constituting the continued LHS baseline at a total budget of 600 solver evaluations. surrogate_guided_300.csv — 300 analyses selected by the surrogate-guided, target-uniform procedure over the engineering band FOS ∈ [0.5, 6.0] (arm C training set). Realized FOS values outside the band are retained. test_uniform_150.csv — Independent test set of 150 configurations, stratified uniformly over the predicted FOS band (10 points per bin, 15 bins), disjoint from all training sets and labeled by independent Spencer analyses. Used as the common evaluation set for all arms. Column descriptions Column Description Unit angle Slope angle (β) ° phi Effective friction angle (φ) ° c Effective cohesion (c) kPa gamma Unit weight (γ) kN/m³ H Slope height m FOS Factor of safety, Spencer's method, HYRCAN — Center_X X-coordinate of the critical slip circle center, measured from the slope toe (origin of the reference system) m Center_Y Y-coordinate of the critical slip circle center, measured from the slope toe (origin of the reference system) m Radius Radius of the critical slip circle m At_Boundary Quality control flag: 1 if the critical surface approached the search boundary within 0.01H before repair, 0 otherwise — slope_length Horizontal run of the slope face m extension Lateral extension of the analysis domain beyond toe and crest: max(1.5H, 0.5·slope_length) m depth Domain depth below the toe: max(H, 0.5·slope_length) m repair_variant Repair step applied to boundary-flagged configurations: blank if none, otherwise depth_2x / both_2x / both_3x in order of increasing domain enlargement — excluded 1 if the configuration was excluded from the final dataset after all repair attempts failed, 0 otherwise — Note on geometry: The reference system has its origin at the slope toe. The X-axis is horizontal (positive toward the crest) and the Y-axis is vertical (positive upward). All geometric quantities (Center_X, Center_Y, Radius, slope_length, extension, depth) are expressed in this coordinate system. Reproducibility The generation pipeline is fully scripted and deterministic (random seed 42). All 300 configurations in the seed, 300 in each of the LHS-continued and surrogate-guided sets, and 150 in the test set were obtained without manual intervention. The code used to generate, quality-control, repair, and select all configurations is available at [GitHub URL].