Optimizing Feedforward neural networks using Krill Herd algorithm for E-mail spam detection

Abstract

Krill Herd is a new optimization technique that was inspired by the herding behavior of real small crustaceans called Krills. The method was developed for continuous optimization problems and has recently been successfully applied to different complex problems. Feedforward neural network has a number of characteristics which make it suitable for solving complex classification problems. The training of the this type of neural networks is considered the most challenging operation. Training neural networks aims to find a nearly global optimal set of connection weights in a relatively short time. In this paper we propose an application of Krill Herd algorithm for training the Feedforward neural network and optimizing its connection weights. The developed neural network will be applied for an E-mail spam detection model. The model will be evaluated and compared to other two popular training algorithms; the Back-propagation algorithm and the Genetic Algorithm. Evaluation results show that the developed training approach using Krill Herd algorithm outperforms the other two algorithms.