Nonlinear ANN Modeling for Predicting Ultimate Strength of CFST Columns

Artificial Neuron Network (ANN) applications are growing strongly in many scientific fields. Especially in the field of structural engineering, ANN has been used for solving many problems, from prediction to classification problems. The datasets performed in those problems are preserved with the original data; general information is normalized if processed. In addition, the structure of the ANN applied is mostly multiple layers of perceptrons (MLPs) without giving precise information as to why MLPs are needed. This study as a contribution is clear to the above problems using ANN. The problem is the ultimate strength prediction problem of the CFST circular columns with a data set of 663 samples, including 6 continuous variables features. The ANN prediction model for the critical intensity of the above column type considers two more issues that have not been clarified, namely i). Does the algorithm used, StandardScaler and MinMaxScaler, for data normalization affect the results of the predictive model? ii). How do ANN structures have only one perceptron layer (Linear ANN – LANN) and ANN structures have multiple perceptron layers (Nonlinear ANN – NANN) with an equal number of units and does the different number of units affect the predicted problem above?