Bat-BP: a new bat based back-propagation algorithm for efficient data classification

Training neural networks particularly back propagation algorithm is a complex task of great importance in the field of supervised learning. One of the nature inspired meta-heuristic Bat algorithm is becoming a popular method in solving many complex optimization problems. Thus, this study investigates the use of Bat algorithm along with backpropagation neural network (BPNN) algorithm in-order to gain optimal weights to solve the local minima problem and also to enhance the convergence rate. This study intends to show the superiority (time performance and quality of solution) of the proposed meta-heuristic Bat-BP algorithm over other more standard neural network training algorithms. The performance of the proposed Bat-BP algorithm is then compared with Artificial Bee Colony using BPNN (ABC-BP), Artificial Bee Colony using Levenberg-Marquardt (ABC-LM) and BPNN algorithm. Classification datasets from UCI machine learning repository are used to train the network. The simulation results show that the efficiency of BPNN training process is highly enhanced when combined with BAT algorithm