Prediction of Earthquake Magnitude by an Improved ABC-MLP

Different algorithms have been used for training neural networks (NNs) such as back propagation (BP), gradient descent (GA), partial swarm optimization (PSO), and ant colony algorithm (ACO). Most of these algorithms focused on NNs weight values, activation functions, and network structures for providing optimal outputs. Ordinary BP is one well known technique which updates the weight values for minimizing error but still it has some drawbacks such as trapping in local minima and slow convergence. Therefore, in this work a population based algorithm called an Improved Artificial Bee Colony (IABC) algorithm is proposed for improving the training process of Multilayer Perceptron (MLP) in order to overcome these issues by optimal weight values. Population based algorithm makes MLP attractive because of the social insect's training algorithm. It investigates the improved weights initialization technique using IABC-MLP. The performance of IABC-MLP is benchmarked against MLP train with the standard BP. The experimental result shows that IABC-MLP performance is better than BP-MLP for earthquake time series data.