Adapting and enhancing mussels wandering optimization algorithm for supervised training of neural networks

Loading...
Thumbnail Image
Date
2015-04
Authors
Abusnaina, Ahmed A. A
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
Developing efficient training method for Neural Networks (NN) in terms of high accuracy is a challenge. In addition, training NN is still highly-time consuming. The Mussels Wandering Optimization (MWO) is a recent metaheuristic optimization algorithm inspired ecologically by mussels movement behavior. The major objective of this thesis is to achieve better performance in terms of convergence training time and classification accuracy for pattern classification by proposing new supervised training methods for Artificial Neural Networks (ANN) based on the MWO algorithm. Increasing the performance, especially in terms of classification accuracy led to an adapted version of the MWO; known as Enhanced-MWO (E-MWO) algorithm. Both the original and the enhanced MWO algorithms are then applied for supervised training of Spiking Neural Networks (SNN). The merits of the proposed methods are validated empirically by using a set of benchmark problems, whereas comparisons are conducted against other common rival training methods. Results show that the MWO-based methods are in average 12 folds faster than other rival methods in terms of training time. In terms of classification accuracy, the MWO is on par with other methods with average accuracy of 80.18%, while the EMWO outperforms other rival methods in 3 problems significantly, and on par with the other methods in classifying the rest 5 problems using ANN with average accuracy of 86.0%. The E-MWO outperforms other methods significantly in classifying 6 problems out of 8 problems using SNN as a classifier.
Description
Please contact ‪Ahmad Abusnaina‬ ‪‬ <abusnaina@ymail.com> for full text content.
Keywords
Artificial Neural Networks (ANN)
Citation