Data-Driven Sensitivity Analysis of Plan Irregularity and Seismic Parameters Affecting Building Response

Understanding the interplay between structural configuration and earthquake characteristics on building response is essential for advancing seismic performance evaluation. While previous studies examined plan irregularities and seismic parameters separately, they often overlooked their interdependent, nonlinear effects on building response. To address this gap, the present study proposes a unified, data-driven methodology that simultaneously incorporates plan irregularity measures and seismic features. Geometric irregularities are represented using four dimensionless indices, including overall aspect ratio (Y/X), two vertical limb ratios (Y₁/Y and Y₂/Y), and the horizontal projection ratio (X₁/Y), while seismic properties such as significant duration (SD), peak ground acceleration (PGA), frequency content (PGA/PGV), and Arias Intensity (AI) are considered. Three modelling approaches: Multiple Linear Regression (MLR), and Random Forest Regression (RFR), Artificial Neural Networks (ANN) were evaluated, with ANN proving the most accurate to capture nonlinear interaction and selected for conducting the sensitivity analysis. The methodology was applied to three plan configurations (C, T, + shaped) under 36 spectrally matched bi-directional ground motions of varying significant duration (SD < 25 s and SD > 25 s). Building responses, including base shear, roof displacement, and roof acceleration, were evaluated using time-history analyses along the two principal directions. Results indicate that seismic parameters, particularly PGA and PGA/PGV, have a substantially stronger influence on building response than geometric irregularities, while the overall aspect ratio (Y/X) is the most influential structural descriptor.