<p>An artificial neural network (ANN) depicts that the information processing mechanism imitating the human brain neuronal system has been successfully applied in many fields like classification, clustering, control, and prediction to name a few. Multilayer perceptron neural networks (MLP NNs) are the ones training of whose neuronal architecture is considered to be the most important and challenging step in the development of neural networks. This difficulty in training is mainly due to a large number of candidate solutions and the different search spaces because of differing datasets involved in the learning stage. Gradient descent and recursive methods have long been used as training measures, which often suffer from limitations of being stagnant in local optimal solutions, improper classification of features, slow rate of convergence and sensitive dependence on initializing variables. To overcome these drawbacks, hybrid approaches (metaheuristic algorithms) using chaos have been used. Inspired from the previously obtained successful results, we have used chaos-based biogeography-based optimizer (BBO) in superior orbit and combined two chaotic maps to train MLP NNs. The proposed methods make use of greater discoverability of search spaces for different datasets and benchmark functions to find global optimal solutions in comparison with the competing algorithms quite significantly.</p>