Comparison of SVR models for predicting the compaction properties of lateritic soils as novel hybrid methods

Abstract

Abstract Soil compaction and related parameters play a pivotal role in the material selection process for earth constructions. Because of time limitations and attention to completion resources, the need to develop models for predicting compaction properties is felt more than ever (i.e., maximum dry unit weight ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>γ</mml:mi> </mml:mrow> <mml:mrow> <mml:mi mathvariant="italic">dmax</mml:mi> </mml:mrow> </mml:msub> </mml:math> ) and optimum moisture content ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>ω</mml:mi> </mml:mrow> <mml:mrow> <mml:mi mathvariant="italic">opt</mml:mi> </mml:mrow> </mml:msub> </mml:math> )) using easily measured index properties. The main purpose of this study is to evaluate the usefulness of conventional SVR, ensembled (additive regression (AD-SVR)), and hybrid SVR (imperialist competitive algorithm (ICA-SVR)) models for predicting the maximum dry unit weight ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>γ</mml:mi> </mml:mrow> <mml:mrow> <mml:mi mathvariant="italic">dmax</mml:mi> </mml:mrow> </mml:msub> </mml:math> ) and optimum water content ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>ω</mml:mi> </mml:mrow> <mml:mrow> <mml:mi mathvariant="italic">opt</mml:mi> </mml:mrow> </mml:msub> </mml:math> ) related to proctor compaction test of lateritic soils, where the used dataset was collected from published literature from specific sites. Results depict that all developed models have acceptable efficiency in predicting the <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>γ</mml:mi> </mml:mrow> <mml:mrow> <mml:mi mathvariant="italic">dmax</mml:mi> </mml:mrow> </mml:msub> </mml:math> with <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msup> <mml:mrow> <mml:mi>R</mml:mi> </mml:mrow> <mml:mrow> <mml:mn>2</mml:mn> </mml:mrow> </mml:msup> </mml:math> larger than 0.844 and 0.719, and the <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>ω</mml:mi> </mml:mrow> <mml:mrow> <mml:mi mathvariant="italic">opt</mml:mi> </mml:mrow> </mml:msub> </mml:math> with <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msup> <mml:mrow> <mml:mi>R</mml:mi> </mml:mrow> <mml:mrow> <mml:mn>2</mml:mn> </mml:mrow> </mml:msup> </mml:math> larger than 0.732 and 0.714 for the training and testing data, respectively, representing the acceptable correlation between observed and predicted values. The results of all SVR models were better than the literature, in which the hybrid ICA-SVR model outperformed other models by considering performance criteria. Overall, the hybrid ICA-SVR algorithm could get the most suitable results than others for estimating both <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>γ</mml:mi> </mml:mrow> <mml:mrow> <mml:mi mathvariant="italic">dmax</mml:mi> </mml:mrow> </mml:msub> </mml:math> and <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>ω</mml:mi> </mml:mrow> <mml:mrow> <mml:mi mathvariant="italic">opt</mml:mi> </mml:mrow> </mml:msub> <mml:mo>.</mml:mo> </mml:math> The ICA-SVR model outperforms other approaches and demonstrates the optimization algorithm’s capacity throughout the optimization process.

Topics & Concepts