MACHINE LEARNING-READY SYNTHETIC DATASET FOR SEISMIC RISK ASSESSMENT OF INDIAN BUILDINGS USING IS 1893:2016

AbstractThe research introduces a new, large-scale synthetic dataset that is intended to facilitate machine learning-based applicationsin Indian infrastructure seismic risk assessment. We utilize India's seismic design code (IS 1893:2016) and Monte Carlosimulation to create 1.25 million physically realistic earthquake-building interaction instances for Delhi (Seismic Zone IV).The framework combines: (1) probabilistic ground motion prediction with regionally calibrated coefficients, (2) dynamicbuilding response through IS 1893 design spectrum analysis, and (3) damage probability estimation with IITK-GSDMAfragility curves. Every scenario contains seven principal characteristics - moment magnitude (4.0-6.5 Mw), rupture distance(lognormal distribution with a median of 50 km), building height (5-30 m, uniform distribution), natural period (from heightusing IS 1893 Clause 7.6.2), peak ground acceleration (PGA), spectral acceleration (Sa), and three probabilities of damagestates (Slight: 0.15g, Moderate: 0.30g, Collapse: 0.60g).The dataset is split into training (937,500 cases) and validation (312,500 cases) sets, organized for immediate use insupervised learning code. Special features are the explicit inclusion of Delhi's alluvium soil amplification (S=1.33) andOrdinary RC frame behavior (R=3.0). Computational reproducibility is facilitated by controlled random seeding (NumPyseed=42), while physical plausibility is ensured through limiting all parameters to ranges found in Indian urban settings. Theopen-access data fill key gaps in labeled seismic data for low- and middle-income countries, especially for buildings designedaccording to Indian codes. Future uses may involve: (1) surrogate ML model training for near-real-time risk assessment, (2)early-warning system development, and (3) retrofit optimization through explainable AI. The approach illustrates how code-conforming engineering analysis can be up-scaled for data-driven decision-making in seismic resilience planning.