Kernel-based support vector regression modeling of fault-controlled coal seam geometry: implications for subsurface stability in Central Iran

Accurate characterization of subsurface seam geometry in fault-controlled basins is essential for reliable mine design and geotechnical planning, particularly under conditions of limited drilling data and pronounced structural complexity. This study develops a robust kernel-based support vector regression (SVR) framework for quantitative modeling of coal seam depth and thickness within the structurally segmented Parvadeh coal basin, Central Iran. The basin is bounded by major fault systems, including the Rostam thrust, Ghori Chai, and Nayband faults, which exert strong control on seam architecture and spatial variability. Drilling data from 26 boreholes were compiled, from which a subset of quality-controlled data was used for model calibration. The proposed framework integrates structural segmentation with data-driven regression, enabling stable predictions under sparse data conditions. Model calibration yielded low prediction errors (SSE ≈ 0.052 for depth and SSE ≈ 1.8 × 10−15 for thickness), indicating strong fitting performance. It should be noted that these values reflect model performance on the calibration dataset. The developed approach provides a quantitative basis for subsurface characterization in fault-controlled coal basins and offers practical relevance for mine planning, overburden load evaluation, and geotechnical risk assessment. The results demonstrate the potential of combining structural analysis with machine learning techniques for improving predictive reliability in data-limited geological environments particularly in tectonically complex and data-constrained mining environments.