Total Dissipated Energy Prediction for Flexure- Dominated Reinforced Concrete Columns via Extreme Gradient Boosting

This study aims to provide an efficient framework for predicting the total dissipated energy level of flexure-dominated reinforced concrete columns via a commonly used machine learning method, extreme gradient boosting. A database including 177 reinforced concrete columns is compiled using open-access databases available in the literature. The proposed framework predicts the target total dissipated energy level depending on seven fundamental features: concrete compressive strength, longitudinal rebar yield strength, shear span-to-depth ratio, longitudinal rebar ratio, transverse rebar volumetric ratio, peak drift ratio, and equivalent damping ratio. Here, a correlation-based quantitative analysis is performed to reveal the effects of selected features on the total dissipated energy capacity. It is observed that the peak drift ratio, yield strength of longitudinal rebars, and concrete compressive strength are the most effective parameters among the other features. K-Fold cross-validation is implemented for the classification process. Validation results show that the three fundamental performance indicators such as the means of correlation of determination, the normalized root mean square error, and the mean absolute percentage error are evaluated as 0.75, 0.38, and 0.33, respectively. The sensitivity of predicted targets to algorithm-based hyperparameters is also investigated. The results of this study are expected to contribute to the energy-based design applications in the scope of predicting the dissipated energy capacity of flexure-dominated reinforced concrete column members.