Lateef_my2nd-repo

Predictive models were developed in this study using soft computing techniques including artificial neural network (ANN), M5P model tree (M5P) and random forest (RF) to predict laterite permeability coefficient and shear strength from index properties (specific gravity, linear shrinkage, liquid limit, plasticity index, fine sand content, and fines content). To achieve this goal, an experimental dataset obtained from laboratory analyses of three hundred laterite samples was divided into a model and a gaging dataset. The model dataset contains two hundred and forty data points, which were divided into training, testing and validation datasets, with 70% for training and 15% each for testing and validation of the proposed models. The gaging dataset contains sixty data points, which were used to evaluate and compare the proposed models’ performances with the existing models using the coefficient of determination, root mean squared error, mean absolute percentage error, and mean absolute error. The proposed models outperformed the existing models and provided satisfactory performance, with ANN models presenting the best performance, followed by RF and then M5P for both the permeability coefficient and the shear strength cases. This implies that the ANN models are the most reliable estimation for the prediction of lateritic soils’ permeability coefficient and shear strength. Thus, they can be used for practical purposes.

本研究采用人工神经网络（artificial neural network, ANN）、M5P模型树（M5P model tree, M5P）与随机森林（random forest, RF）等软计算技术开发预测模型，旨在通过红土的指标特性（index properties）——包括比重（specific gravity）、线缩率（linear shrinkage）、液限（liquid limit）、塑性指数（plasticity index）、细砂含量（fine sand content）及细粒含量（fines content）——预测其渗透系数（permeability coefficient）与抗剪强度（shear strength）。
为达成该研究目标，本研究将基于300组红土试样实验室分析得到的实验数据集划分为建模数据集与独立测试数据集。其中，建模数据集包含240组数据点，进一步划分为训练集、测试集与验证集，训练集占比70%，测试集与验证集各占15%，用于构建所提预测模型。独立测试数据集包含60组数据点，用于评估并对比所提模型与现有模型的性能，评估指标包括决定系数（coefficient of determination）、均方根误差（root mean squared error）、平均绝对百分比误差（mean absolute percentage error）与平均绝对误差（mean absolute error）。
实验结果显示，所提模型的性能均优于现有模型，且表现良好；其中人工神经网络模型性能最优，其次为随机森林，M5P模型位列第三，该结论在渗透系数与抗剪强度两类预测任务中均成立。这表明人工神经网络模型是预测红土渗透系数与抗剪强度的最可靠估算方法，可应用于实际工程场景。