Prediction of SPT value based on CPT data and soil properties using ANN with and without normalization

Abstract

Artificial neural networks (ANN) are now widely used and are becoming popular among researchers, especially in the geotechnical field. In general, data normalization is carried out to make ANN whose range is in accordance with the activation function used. Other studies have tried to create an ANN without normalizing the data and ANN is considered capable of making predictions. In this study, a comparison of ANN with and without data normalization was carried out in predicting SPT values based on CPT data and soil physical properties on cohesive soils. The input data used in this study are the value of tip resistance, sleeve resistance, effective soil overburden pressure, liquid limit, plastic limit and percentage of sand, silt and clay. The results showed that the ANN was able to make predictions effectively both on networks with and without data normalization. In this study, it was found that the ANN without data normalization showed a smaller error value than the ANN with data normalization. In the network model without data normalization, RMSE values were 3.024, MAE 1.822, R2 0.952 on the training data and RMSE 2.163, MAE 1.233 and R2 0.976 on the test data. Whereas in the ANN with data normalization, the RMSE values were 3.441, MAE 2.318, R2 0.936 in the training data and RMSE 2.785, MAE 2.085 and R2 0.963 in the test data. ANN with normalization provides a simpler architecture, which only requires 1 hidden layer compared to ANN without normalization which requires 2 hidden layer architecture.

Sensitivity Analysis Based on Physical Properties to Permeability Coefficient of Cohesive Soil Using Artificial Neural Network

“…The forcasting models carried out are then validated using a number of indicators. Commonly used indicators are the mean of the Root Mean Square Error (RMSE), Mean Abosulute Percentage Error (MAPE), Mean Absolute Error (MAE), Mean Absolute Deviation (MAD), Mean Square Error (MSE), tracking signal, and stability testing (Fernando et al, 2021).…”

A correlation based on pressuremeter, SPT and CPT tests for characterizing of coastal alluvium: A study for phase 14 South Pars, Iran

“…The test results are usually used for the calculation of pressuremeter modulus (E m ), limit pressure (P L ), cone resistance (q c ), and sub-grade reaction modulus (K s ) parameters. Several researchers used CPT, SPT, and PMT results to correlate the information and formulate the empirical relation between the CPT-SPT-PMT by using regression analysis to prepare more accurate results with relatively high coefficient of determination (R 2 ) values [4] , [5] . Table 1 provides information about the empirical relations that are estimated for E m based on SPT-CPT values.…”

Analyzing Geotechnical Characteristics of Soils in Erbil via GIS and ANNs

“…The network architecture was chosen using hidden layers and varying the number of neurons in the hidden layer. The relation number of neurons in the hidden layer is between 15 and 18 according to previous researches [29,30]. The network performance that has the smallest error and the correlation coefficient value that is proximate to 1 is most suitable for data predictions.…”