The Artificial Neural Network Design Procedure
The design and deployment of the Artificial Neural Network (ANN) for this research work was divided into six (6) major stages, namely: data acquisition stage; feature selection and data normalization stage, ANN architecture optimization stage, the ANN algorithm optimization stage, ANN initialization and training stage, testing, validation and deployment stage. Feed Forward neural network with Levenberg-Marquardt back propagation ANN model of MATLAB training was used for the computation of data and to determine the best model. The neural network tool box of MATLAB was used for necessary computation required for the development of the models. The Coefficient of Correlation (R) and the Root Mean Square Error (RMSE), were employed to determine the degree of correlation between the target of the soft computing models and their eventual outputs. The six input variables were cement (%), Rice Husk Ash (RHA) (%), Liquid Limit (LL) (%), Plasticity Index (PI) (%), Maximum Dry Density (MDD) (Kg/m3) , and OMC (%), while CBR Unsoaked (%) and CBR Soaked (%) were the output variables. From the experimental results, 322 set of soil data were obtained, the data were subdivided into 70% for training, 15% for testing and 15% for validation. Table 1 shows details of the components of the ANN model.
Table 1: Details of components of the ANN model.