Skip to Main content Skip to Navigation
Journal articles

Comparison and Evaluation of Artificial Neural Network (ANN) Training Algorithms in Predicting Soil Type Classification

Abstract : The present study uses different ANN training algorithms to predict soil type classification and evaluates the output of their training. Studies were done on the soil of Shahrekord (Iran), using a database consists of 120 soil samples. The used data includes the results of standard impact and penetration tests, classification and Atterberg limits. Because of diverse training algorithms in neural networks, the selection of the optimal training function can increase the accuracy of network predictions. For this purpose, several training functions and algorithms were used in the ANN modeling. To evaluate the performance of proposed models, relations of coefficient of residual mass (CRM), coefficient of determination (COD) and root mean square error (RMSE) were used. The comparison and evaluation of training with various algorithms show that the Levenberg-Marquardt training function with very high accuracy in network training is the optimal training function used in the prediction of geotechnical parameters of soil, including soil type.
Complete list of metadatas

https://hal.univ-lorraine.fr/hal-02541242
Contributor : Rasool Mehdizadeh <>
Submitted on : Monday, April 13, 2020 - 1:09:15 PM
Last modification on : Wednesday, November 4, 2020 - 1:56:03 PM

Identifiers

  • HAL Id : hal-02541242, version 1

Citation

Hamidreza Hassannejad, Mohammad Sirus Pakbaz, Rasool Mehdizadeh. Comparison and Evaluation of Artificial Neural Network (ANN) Training Algorithms in Predicting Soil Type Classification. Pharmacology and Life Sciences Bull. Env.Pharmacol. Life Sci, 2015, 4, pp.212 - 218. ⟨hal-02541242⟩

Share

Metrics

Record views

30