Comparison of linear regression, deep learning and random forest algorithms for predicting ultimate load capacity of nonlinear inelastic analysis of steel frames

Abstract

The rapid and powerful development of computer science and computing power in recent decades has promoted the application of advanced analytical methods to engineering design problems in general and steel frame design practice in particular. One of the possible and popular directions is to apply machine learning algorithms to predict the behavior of steel frame structures in nonlinear inelastic analysis. This shows obvious advantages such as speeding up the decision-making process, reducing error rates, and increasing computational efficiency. In this paper, the effectiveness of three popular machine learning algorithms is studied for the prediction of the load-carrying capacity of steel frames including Linear Regression, Deep Learning, and Random Forest. A numerical example surveying a 5-span 14-story planar steel frame is considered. An advanced nonlinear inelastic analysis is performed for the steel frame to generate training datasets to minimize analysis time. The input variables of the problem are the geometrical characteristics of the beam and column cross-section selected from the available list. The performance of the machine learning algorithms was evaluated using error indexes including mean square error (MSE), and coefficient of determination (R2) and the results showed that the random forest method is the most effective among the three machine learning methods selected.

Keyword: Steel frame; nonlinear inelastic analysis; machine learning;