Sonar Dataset Classification using Multi-Layer Perceptron Neural Network Based on Dragonfly and Moth Algorithms

One of the most complex areas of sonar research is the classification and recognition of the real target from the liar. Multi-layer perceptron neural networks (NNs) are the most popular and fastest classifier in this area. Train of these networks in remarkable in recent years. Classical algorithms for the training of NNs include: recursive methods, gradient descent, and Newton, etc. Some disadvantages of these methods are improper accuracy, trapping in local optimum, and low convergence rate. In recent years, metaheuristic algorithms combined for the training of NNs are proposed for dominating these defects. In this paper, two new meta-heuristic algorithms are used based on mimicking from animals (dragonfly and moth) for the training of NNs. Simulated results on Iris and Sejnowski datasets are shown Moth–Flame classification rate is 88% and has 30% improvement rather than old methods.