Optimizing Input variables for the Artificial Neural Network Model Using Genetic Algorithm

Abstract

ANN models are known to give satisfactory results when it comes to simulation and ultimate prediction of events irrespective of the process as far as there are sufficient historical data which the machine learning technique can simulate. However, the process can be modified such that the ANN model can even give better results than it is giving presently; Genetic Algorithm Technique in WinGAmma software was utilized in analysing the dataset in order to select the most relevant combination of input variables from the dataset which will produce better performance of the ANN model and the output was compared with the normal ANN model which utilized all the input variables in the dataset and the GA-ANN model outperformed the ANN model with all input variables, which goes on the show the GA technique is very reliable and a welcome technique in selecting input variables for the optimal performance of the ANN model in simulation and prediction of events. The event simulated in this work is the Daily suspended sediment load with several input variables ranging from present and antecedent parameters of Discharge, Sediment Discharge and Suspended Sediment load. Although the result of the ANN model with all 7 input variables was satisfactory, but the GA-ANN model gave a much better result with lesser but more relevant input variables as selected by the GA technique.