Feature selections and optimizable classification learners for detecting failure modes of rectangular reinforced concrete columns

One of the key steps in the framework of seismic risk and strengthening evaluations of existing reinforced concrete (RC) bridges, frames, or buildings is the identification of failure modes of RC columns. This paper deals with an efficient method based on machine learning techniques to classify failure modes of rectangular RC columns due to lateral loadings. In this regard, various classification learners such as decision tree, discriminant analysis, naive Bayes, k-nearest neighbor, support vector machine, neural network, and ensemble are employed with an adequate collected dataset of 310 quasi-static cyclic tests. Based on feature selection analyses of various methods, five parameters are used as the input for the model training, and the output is one among three failure modes of the columns including flexure, flexure-shear, and shear. Optimized classifiers are also obtained using the Bayesian optimization scheme on a range of hyperparameters to improve the performance capacity of the models. As a result of the cross-validation on both training and separate test sets, which is in terms of the confusion matrix, the support vector machine, ensemble, and k-nearest neighbor classifiers all exhibit very high classification performances with accuracy percentiles of more than 94%.