Prediction of Consolidation Parameters Using Artificial Neural Network Model & Regression Analysis: A Case study of Lege Tafo Area

The Coefficient of compression index (Cc) and Coefficient of recompression index (Cr) are
important compressibility parameters to work out the consolidation settlement for fine-grained soil
layers. These parameters can be obtained in the laboratory by conducting one-dimensional
oedometer test on undisturbed samples. However, consolidation tests require high quality of
undisturbed soil sample, costly and time-consuming. Therefore, it is important to determine
consolidation parameters based on basic index soil property tests. In this research, it is attempted to
evaluate and examine the artificial neural network (ANN) technique and multiple regression to
estimates the soil compressibility parameters (Cc and Cr) based on conveniently measurable soil
index properties. According to the ASTM and AASHTO, the soil in the study area is categorized as
highly plastic clays. An artificial neural network (ANN) of two-layer feed forward back propagation,
having input variables, hidden layer and target nodes algorithm have been applied to predict Cc and
Cr index. The Levenberg-Marquardt method was used as the training algorithm and the hyperbolic
tangent function as an activation function for both the hidden and output layers. The results showed
that 4-5-1architecture is found to be quite satisfactory in predicting the Cc with a coefficient of
correlation R2 of 0.89 and root mean square error, RMSE of 0.0207, whereas for Cr, 4-4-1
architectures performs better with R2 of 0.88 and RMSE of 0.019. In the multiple regression
analysis, R2 of 0.516 and 0.321 with RMSE of 0.063 and 0.0353 were found for Cc and Cr
respectively. By comparing the values of the correlation coefficient, R2 and root mean square error,
RMSE of the two methods used, it was revealed that ANN has the least error and the most
Therefore, this study recommends estimating the compression index (Cc) and recompression index
(Cr) from the newly developed equations from ANN and regression analysis.