Title: Feedforward neural network based on ensemble evolutionary algorithm with self-adaptive strategy and parameter for intrusion detection

Authors: Yu Xue; Tao Tang; Alex X. Liu

Addresses: School of Computer and Software, Nanjing University of Information Science and Technology, Pukou, Nanjing, China ' School of Computer and Software, Nanjing University of Information Science and Technology, Pukou, Nanjing, China ' Department of Computer Science and Engineering, Michigan State University, East Lansing, USA

Abstract: The applications of Wireless Sensor Networks (WSNs) have driven the development of many fields, and the information security is an important issue in the WSNs. Intrusion Detection System (IDS) is a kind of effective system for identifying network attacks and protecting data security, which has a great significance for WSNs. In this paper, we transform the intrusion detection problem into a classification problem and introduce the Feedforward Neural Network (FNN) classifier to solve it. At the same time, we propose a novel Self-adaptive Parameter and Strategy Differential Particle Swarm (SPS-DPS) optimisation algorithm to find the optimal weights for the FNN. Experiments are performed on eight data sets which are constructed based on the intrusion detection data set KDDCUP99, and the results are compared with five Evolutionary Computation (EC) methods. The results show that the SPS-DPS based FNN method can solve the intrusion detection problem well compared with the other methods.

Keywords: WSN; wireless sensor network; intrusion detection; FNN; feedforward neural network; classification; evolutionary computation; particle swarm optimisation; differential evolution; self-adaptive strategy; parameter.

DOI: 10.1504/IJWMC.2019.101438

International Journal of Wireless and Mobile Computing, 2019 Vol.17 No.2, pp.202 - 211

Received: 18 Feb 2019
Accepted: 27 Apr 2019
Published online: 07 Aug 2019 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article