Premium
Modeling of maximum dry density and optimum moisture content of stabilized soil using artificial neural networks
Author(s) -
Hossein Alavi Amir,
Hossein Gandomi Amir,
Mollahassani Ali,
Akbar Heshmati Ali,
Rashed Azadeh
Publication year - 2010
Publication title -
journal of plant nutrition and soil science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.644
H-Index - 87
eISSN - 1522-2624
pISSN - 1436-8730
DOI - 10.1002/jpln.200800233
Subject(s) - artificial neural network , multilayer perceptron , water content , parametric statistics , perceptron , range (aeronautics) , computer science , biological system , mathematics , soil science , artificial intelligence , statistics , environmental science , engineering , geotechnical engineering , biology , aerospace engineering
This study considers the use of artificial neural networks (ANNs) to predict the maximum dry density (MDD) and optimum moisture content (OMC) of soil‐stabilizer mix. Multilayer perceptron (MLP), one of the most widely used ANN architectures in the literature, is utilized to construct comprehensive and accurate models relating the MDD and OMC of stabilized soil to the properties of natural soil such as particle‐size distribution, plasticity, linear shrinkage, and the type and quantity of stabilizing additives. Five ANN models are constructed using different combinations of the input parameters. Two separate sets of ANN prediction models, one for MDD and the other for OMC, and also a combined ANN model for multiple outputs are developed using the potentially influential input parameters. Relative‐importance values of various inputs of the models are calculated to determine the significance of each of the predictor variables to MDD and OMC. Inferring the most relevant input parameters based on Garson's algorithm, modified ANN models are separately developed for MDD and OMC. The modified ANN models are utilized to introduce explicit formulations of MDD and OMC. A parametric study is also conducted to evaluate the sensitivity of MDD and OMC due to the variation of the most influencing input parameters. A comprehensive set of data including a wide range of soil types obtained from the previously published stabilization test results is used for training and testing the prediction models. The performance of ANN‐based models is subsequently analyzed and compared in detail. The results demonstrate that the accuracy of the proposed models is satisfactory as compared to the experimental results.