Machine-Learning-Driven Approaches for Assessment, Delegation, and Optimization of Multi-Floor Building

Abstract

This study presents a novel integrated framework for the structural analysis and optimization of multi-floor buildings by combining validated theoretical models with machine learning and evolutionary algorithms. The proposed Process–Action–Response System (PARS-Solution) accurately computes key structural responses—such as deformations, shear forces, and bending moments—based on eleven critical design parameters ($P_1$ to $P_{11}$). The significance of this research lies in its ability to automate and accelerate complex structural analysis using Adaptive Neuro-Fuzzy Inference Systems (ANFISs), achieving an average error of less than 2% in multi-variable prediction scenarios. The results were compared against reference calculations and ETABS simulations to validate its effectiveness, demonstrating deviations of less than 3%. The methodology combines MATLAB-based coding, interpolation from verified reference diagrams, and iterative stiffness adjustment across floors, offering transparency and accuracy. Optimization is performed using Multi-Objective Particle Swarm Optimization (MOPSO), enabling efficient exploration of Pareto-optimal solutions that balance deformation and material usage. Extensive parametric studies reveal the dominant impact of core wall dimensions and floor number on structural efficiency, while the application of stiffness reduction factors (e.g., $P_{11}$) proves effective in reducing material without compromising performance. This hybrid approach enables the delegation of labor-intensive calculations to a trained ANFIS model and supports rapid pre-validation of structural configurations in early design phases. As such, the framework offers a powerful data-driven tool for engineers seeking optimal, lightweight, and high-performance solutions in high-rise building design.