Energy Efficient Secure Communication Using Machine Learning and Trust Model

Abstract

In general, a sink and various tiny sensor nodes are included in Wireless sensor networks (WSNs). Overall performance and security of WSNs are degraded due to limited resources and hostile environments, broadcasted and untrusted transmissions, unprotected and free communications and nodes misbehaviour due to malicious selfishness or compromised intentions. Against various attacks, WSNs’ security is also degraded. To overcome those issues in recent work used Guard against Trust Management with Neural Network classifier (GATE-NN) and Improved Ant Colony Optimization (IACO) for optimal selection of window length. In a sinkhole attack, for detecting intruder, a Neural Network (NN) algorithm is suggested in this work. Data consistency is analysed at first, for computing suspected nodes group. Then, network flow information is checked for effective recognition of intruder. However existing work failed to focus on energy efficiency of the nodes in the network and it may lead to the network failure. And still have some accuracy   issues   in   sinkhole   attack detection. To overcome those issues this work first introduced  a probability based fuzzy  c means clustering model for network node clustering and then cluster heads will be selected based on uniform distribution based fuzzy neural network. And additionally define a new attack detection system which suitable to the properties of WSNs. The proposed work is implemented in the Guard Against Trust Management with weighted mean support vector machine classifier (GATE-WMSVM) and Improved Ant Colony Optimization (IACO) for optimal selection of window length. System resources requirement is minimized by considering dynamic timing window and trust domain based on positive integer. In order to improve in detecting sinkhole attack this work uses  support vector machine (SVM) algorithm.Data consistency is analysed at first, for computing suspected nodes group in this work. Then, network flow information is checked for effective recognition of intruder.For every node’s malicious and good behaviour, penalty and reward policies are used in this direct trust establishment technique. This makes highly realistic trust computation.