Applying Kernel Principal Component Analysis for Enhanced Multivariable Regression Modeling of Rubberized Concrete Properties

Abstract

Nowadays, intensive investigations on the potential of rubberized concrete as a structural material are available in the literature due to its promising characteristic in terms of sustainability, ductility, high absorption of energy, and enhanced vibration behavior. These works have initiated the introduction of multiple estimation methods to rapidly evaluate its properties. Over the last few decades, hundreds of researches have been conducted to assess different mathematical approaches for modeling concrete's properties. Among them, multivariable regression analysis is considered one of the simplest multipurpose options. However, its adaptation is sometimes inefficient, and its results are not highly accurate. This study proposes an enhancement to the estimation technique for rubberized concrete's mechanical characteristics by conducting a data reconstruction using kernel principal component analysis and then utilizing multivariable regression to develop the prediction models. The importance of the kernel principal component analysis comes from its capability to transform the input parameters to visualize the dataset's information described by multiple inter-correlated quantitative variables so that it discovers underlying patterns. In general, the research results have shown that the performance of the proposed method can significantly overcome that of the conventional regression techniques with as high as 80% reduction in the root-mean-square error in the case of compressive strength.