Modified Particle Swarm Optimization (Mpso) Feature Selection and Replicated Neural Network (Rnn) Based Intrusion Detection System (Ids)

Abstract

Wireless Sensor Networks (WSNs) is a large-scale ad hoc network with the property of
adequate availability, including the all scopes of communication applications, like health care,
home automation, remote observations, obstacles detection, etc. WSN is exposed to many
security threats due to its limitations, broadcast nature and unattended environment. Intrusion
Detection System (IDS) allows detecting suspicious or abnormal activities and triggers an
alarm when an intrusion occurs. The implementation of IDSs for WSNs is more difficult than
other systems because sensor nodes are usually designed to be tiny and cheap, and they do
not have enough hardware resources. Thus it is required to implement the faster and overhead
reduced IDS for ensuring the increased attack detection rate. In this paper, Pattern Matching
aware Replicated Neural Network based Intrusion Detection System (PM-RNN-IDS) is
introduced for the accurate and faster IDS rate. Firstly, pre-processing redundant record
detection is done by using Firefly Algorithm (FA) to eliminate redundant record set and then
Enhanced k Nearest Neighbour (EkNN) based missing imputation is introduced to handle
missing value. Then most important features in the dataset is selected and removed from the
original feature set by Modified Particle Swarm Optimization (MPSO). Finally intrusion
detection is carried out using Replicator Neural Networks (RNNs). In the experimentation
evaluation, Network Simulator (NS-2) simulator is used to evaluate the performance of the
proposed RNNs system, and existing classifier with 100 nodes by a 100 × 100 meters area.
The results are evaluated using the metrics like False Alarm Rate (FAR), Detection Rate
(DR), and Accuracy (ACC).