The buckling behavior of columns, as the most influential members regarding the stability of structures, has been a long-standing field of interest. Moreover, due to the conservative attitude of classical theories, the ultimate buckling load obtained through these recommendations is mainly considered an underestimation of its actual value. This paper hence aims to develop practical metaheuristic-trained Artificial Neural Networks to predict the ultimate buckling load of High Strength Steel columns. Therefore, initially, the ABAQUS Finite Element Model was verified with experimental results and showed 99.82% accuracy; then, nonlinear finite element analyses for 114 Models were conducted to be utilized as datasets to develop ANNs. To do so, the section height, flange width, web thickness, flange thickness, and steel yield strength were considered the input variables, while the ultimate buckling load was assumed to be the only output variable. ANN models are typically trained through algorithms such as LM, BR, and SCG; meanwhile, the novelty of this work is to take advantage of metaheuristics to optimize weights and biases of ANNs. In this regard, four metaheuristic algorithms, namely Particle Swarm Optimization, Colliding Body Optimization, and their developed version by applying Genetic Algorithm, Particle Swarm Optimization-Genetic Algorithm, and Colliding Body Optimization-Genetic Algorithm, have been used to train ANNs efficiently. Besides, their performances were evaluated using statistical relations like Coefficient of Cor-relation and Mean Square Error. The results illustrate that the trained models could accurately predict the ul-timate buckling loads up to 99.8%, demonstrating the efficiency and precision of the present work.
