Feature selection approach for failure mode detection of reinforced concrete bridge columns

Selecting optimal input variables for machine learning (ML) algorithms is essential for any model outputs. This study presented a feature selection-based approach for determining the optimal input parameters for classifying reinforced concrete columns failure modes. The comprehensive datasets of 311 reinforced columns involving different parameters were collected from the previous studies. The Pearson correlation (PC) and mutual information (MI) techniques were used to test input variables' linear and nonlinear relevance to the outputs. In addition, minimum redundancy maximum relevance (mRMR) algorithms were employed to select and rank the relevance of eleven input variables for the model outputs. i.e., flexural (F), flexural-shear (F-S), and shear (S) failure modes using predictor importance score. Three different classification al-gorithms, artificial neural networks (ANN), Decision Tree (DT), and Naive Bayes (NB), were used to analyze five different models, M1 to M5, developed using different combinations of the selected input variables. The aspect ratio, longitudinal rebar index, transverse rebar index, and axial load ratio are the optimal input parameters that classify the failure mode reinforced concrete column.

Automated summary

This study presented a feature selection-based approach for determining the optimal input parameters for classifying reinforced concrete columns failure modes. The Pearson correlation and mutual information techniques were used to test the relevance of input variables to the outputs, and the minimum redundancy maximum relevance algorithm was employed to select the important input variables. The aspect ratio, longitudinal rebar index, transverse rebar index, and axial load ratio were identified as the optimal input parameters.