Improved Selection Method for Evolutionary Artificial Neural Network Design

this paper improves the role of adaptive nature of new Evolutionary Algorithm (EA) [19] in designing Artificial Neural Network (ANN) using the proper selection mechanism. The proposed EA has been used for two purposes. One is generalization of architecture. In this, the optimal adaptive architecture is achieved by using evolutionary crossover and mutation. The adaptive strategy increased in the stage of selection process. This algorithm used the tournament selection method with minimum hamming distance. Unlike most previous studies, proposed EA puts emphasis on autonomous functioning in the design process of ANNs. The mathematical frame work is discussed in [19]. The proposed EA has been tested on a number of benchmark problems in machine learning and ANNs, including breast cancer, diabetes, heart problems and for time complexity N-Bit Parity is used. The experimental results show that proposed EA can design compact ANN architectures with good generalization ability, compared to other algorithms with good time complexity.