Predicting Resilient Modulus of Subgrade Soil Using Deep Learning Technique

Abstract

The description of the subgrade hardness is frequently dependent on the application of the MR (resilient modulus) of the subgrade soils, which acts as an essential consideration. This study is based on the use of Deep Learning technique (ANN) and machine learning equations, for determination the MR of soil uses in subgrade layers which is an effective and reliable manner. According to AASSTO, A-4 group of soil is selected for this research. The research database has been developed with 119 experimental test results. MR works as an output parameter, including Liquid Limit, Plastic Limit, Optimum Moisture Content, Maximum Dry Density, and California Bearing Ratio as input parameters. Machine learning algorithm and ANN model is created using Python programming language in Google Colab. Multiple machine learning (ML) techniques, such as linear regression, lasso regression, ridge regression, and K-nearest neighbors (KNN), are used along with the deep learning (DL) technique of BPNN (backpropagation neural network) and optimizer (Adam) is used for predicting MR. The accuracy of the predicted data is determined by evaluation metrics like MSE (mean squared error), R2(R-squared), MAE (mean absolute error), and MAPE (mean absolute percentage error). BPNN, utilizing Adam optimizers, exhibits high precision in predicting MR results, as seen by its high statistical relation values of R2 and minimal error rates.