Scholarlink

<p>Several machine learning (ML) techniques are used in this work to forecast the relative density (RD) of the soil in Katihar, Bihar, India. Although infrastructure development is crucial to any place’s progress, Katihar is still in its early phases of development. Engineers can use the relative density of the soil to determine the bearing capacity, settlement, liquefaction assessment, and ensure the stability of the structure with the help of relative density of soil. The relative density of soil is predicted using three machine learning (ML) models, namely <i>k</i>-nearest neighbour (KNN), deep neural network (DNN) and polynomial regression (PR). Hundred datasets are used for this study, which is performed in the laboratory. The dataset comprises eight input parameters, namely bulk unit weight of soil (<i>γ</i><sub>b</sub>), water content, specific gravity of soil solids, percentage of sand, percentage of fines (% <i>F</i>), dry unit weight of soil, dry unit weight of soil in the loosest state, and dry unit weight of the soil in the densest state and the relative density of the soil as the output. Numerous statistical performance parameters, rank analysis, <i>R</i>-curve, reliability analysis, error matrix, and Taylor diagram are used to assess the model’s accuracy. Based on these assessment parameters, it has been observed that the PR model performs better than the other two models in terms of predicting RD. The PR model performs better due to its greater <i>R</i><sup>2</sup> = 1 and lowest RMSE = 0.000 in the training phase and higher <i>R</i><sup>2</sup> = 0.994 and lower RMSE = 0.001 in the testing phase. The first-order second moment approach is also used to calculate the reliability index. Sensitivity analysis is also conducted to assess the impact of each input parameter on the output (RD of soil). Based on all the findings, PR is the most accurate method for predicting RD of soil, followed by DNN and KNN. The outcome of this study provides engineers and researchers important insights into the applicability of PR models in predicting the RD of soil, with substantial consequences for the field of geotechnical engineering.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Experimental and Computational Response of Relative Density of Soil of Katihar, India

  • 期刊论文

  • 发表日期：

    2025年2月7日

查看全文
  • Rashid Mustafa

摘要

Several machine learning (ML) techniques are used in this work to forecast the relative density (RD) of the soil in Katihar, Bihar, India. Although infrastructure development is crucial to any place’s progress, Katihar is still in its early phases of development. Engineers can use the relative density of the soil to determine the bearing capacity, settlement, liquefaction assessment, and ensure the stability of the structure with the help of relative density of soil. The relative density of soil is predicted using three machine learning (ML) models, namely k-nearest neighbour (KNN), deep neural network (DNN) and polynomial regression (PR). Hundred datasets are used for this study, which is performed in the laboratory. The dataset comprises eight input parameters, namely bulk unit weight of soil (γb), water content, specific gravity of soil solids, percentage of sand, percentage of fines (% F), dry unit weight of soil, dry unit weight of soil in the loosest state, and dry unit weight of the soil in the densest state and the relative density of the soil as the output. Numerous statistical performance parameters, rank analysis, R-curve, reliability analysis, error matrix, and Taylor diagram are used to assess the model’s accuracy. Based on these assessment parameters, it has been observed that the PR model performs better than the other two models in terms of predicting RD. The PR model performs better due to its greater R2 = 1 and lowest RMSE = 0.000 in the training phase and higher R2 = 0.994 and lower RMSE = 0.001 in the testing phase. The first-order second moment approach is also used to calculate the reliability index. Sensitivity analysis is also conducted to assess the impact of each input parameter on the output (RD of soil). Based on all the findings, PR is the most accurate method for predicting RD of soil, followed by DNN and KNN. The outcome of this study provides engineers and researchers important insights into the applicability of PR models in predicting the RD of soil, with substantial consequences for the field of geotechnical engineering.