Optimization of Neural Network Using Improved Bat Algorithm for Data Classification

Artificial neural network (ANN) contributes significant advantages in medical research. In the field of medical imaging and health care application, ANN performs the significant role by applying the neural networks. However, nowadays vast abundance of efforts is concentrated on the evolution of neural networks for application purposes such as data compression, pattern recognition, optimization, and classification. Bat algorithm (BA) is a natureinspired metaheuristic algorithm, which is widely used to solve the real world global optimization problem. One of the major issues faced by the BA is its frequent capture in local optima while handling the complex real-world problems. In this study, a new variant of BA, named as improved bat algorithm (IBA) has been proposed. The proposed approach modifies the standard BA by enhancing its exploitation capabilities and avoids escaping from local minima. Artificial neural network (ANN) has a wide variety of practice for the solution of optimization problems in the area of data classification. The process of training of ANN is a complex continuous optimization task and has great importance in supervised learning. Back propagation algorithm (BPA) is a famous ANN traditional training approach. Since this classical training technique has the drawbacks like stuck in the local minima and maximum number of iterations required. During weight optimization, the paper intended to demonstrate a detailed comparative performance analysis for the training of feed forward neural network (FFNN) of the proposed technique (IBANN) over other gradient decent algorithms and population based approaches on the different data sets taken from UCI repository. The experimental results show that IBANN outperforms BPA and BPA with momentum, training of ANN with BA (BANN) and particle swarm optimization (PSONN) in terms of converging speed and better accuracy. The simulation depicts the superiority of IBA and proves it best substitute to traditional training approach for ANN for the solution of classification problems. This improved version of BA algorithm can also be used for the medical images analysis.