Bio-Inspired Optimization for Medical Data Mining

13Optimizing Artificial Neural‑Network Using Genetic Algorithm

Bhavy Pratap and Sulabh Bansal*

Department of Information Technology, Manipal University Jaipur, Jaipur, Rajasthan, India

Abstract

Artificial neural networks (ANNs) is a processing system which is adaptive and nonlinear in nature. One has to select the suitable number of neurons to perform ANN as insufficient number of neurons can lead to poor approximation, whereas excessive neurons can result in overfitting issues. In order to improve the accuracy and simplify the structure of neural networks, a variety of optimization techniques and algorithms has been developed. These techniques and algorithms help to train neural networks more efficiently and effectively, leading to better performance. One of those algorithms is Genetic Algorithm (GA). GA is random-search algorithm, which takes the nature selection and evolution process into consideration. Both NNs and GA are similar techniques. The similarity between ANNs and GAs inspired us to combine the two approaches to see if a GA could be used to train neural networks with high accuracy. A GA is a search algorithm that uses principles inspired by biological evolution, such as mutation, crossover, and selection. First, the “selection” mechanism chooses best parents from a gene pool, which is essentially an array containing the best matrix of weights, in order to obtain the most favourable genes. Second, the crossover process is implemented by selecting two genes randomly ...