Parametric Analysis and Explainable AI-Based Interpretation of the Seismic Response of Timber Modular Buildings

Timber modular construction has attracted increasing attention as a means of improving construction productivity and reducing on-site work. However, its application in Japan remains limited, and the effects of building-scale configurations and connection conditions on the seismic response of multi-story timber modular buildings are not fully understood. This study investigated these effects by combining parametric nonlinear time-history analysis with explainable artificial intelligence (XAI). Three-dimensional analytical models consisting of CLT wall panels, glulam beams, and nonlinear springs representing structural connections were developed. A full-factorial design comprising 648 analysis cases was conducted by varying six parameters: number of stories, number of horizontally arranged modules, structural configuration, presence of wall-to-wall shear connections, wall-end connection type, and input ground motion. Four inter-story drift indices were evaluated. Gradient Boosting models trained on the numerical results achieved R2 values of 0.910–0.954, and SHapley Additive exPlanations (SHAP) identified the number of stories and structural configuration as the dominant parameters. Wall-dominant configurations tended to reduce deformation in one- and two-story buildings, whereas this tendency diminished or reversed in buildings with three or more stories, indicating a strong height-dependent interaction between structural configuration and seismic response. These findings provide useful insight for rational seismic design of timber modular buildings.