GA and PSO hybrid algorithm for ANN training with application in Medical Diagnosis

Since the diagnosis of a disorder in patients is very important for proper treatment, so it is essential that the results of these diagnostic tests are properly categorized. ANNs have the computational capability of efficiently mapping the data collected from medical tests to the results. In this paper, it is analyzed that how hybridizing the training process of ANNs with Evolutionary algorithms such as PSO and GA can enhance the mapping capability of ANNs and overcome the shortcomings of Gradient Descent training algorithm. When compared to Gradient Descent, the proposed GA-PSO-BP hybrid training algorithm, demonstrated an average increment of 5% in the average testing accuracies. While the PSO-BP hybrid training algorithm displayed an average improvement of 4% as compared to Gradient Descent.