Implementation of Trust Model for Social Internet of Things Using Centrality Feed Forward Networks

Abstract

Recent advances of Internet of Things (IoT) lead to the most promising paradigm called Social Internet of Things (SIoT). These techniques are considered the strong amalgamation of the social networking features with the IoT objects. These networks are characterized by facilitating the IoT objects to establish the social networking between each other. In SIoT, an interconnection of networks formed by considering the important features such a object-object interactions, social relationship, reliable recommendations and mandates the careful attention towards the strong trustworthy connections. Hence Intelligent Trust Model Identification Model System (ITMIS) is required for the SIoT networks to identify the misbehaving objects by selecting only the reliable, credible and trustworthy objects before relying on the services provided by them. However, existing frameworks truly relay on the conventional approaches that is based on linear relationship between the inputs and outputs. These methods may lead to the high misclassification ratio of selecting the untrusted devices that even may cause untrust process in an application. To overcome this problem, in the proposed research work the novel ITMIS which ensembles the non-linear centralities relationships with the architecture Extreme Feed Forward Neural Networks (EFFNN) with the combination of hybrid algorithm of Long Short Term Memory (LSTM) and Gated Recuurent Unit (GRU) for the better accuracy compare with the existing model. The proposed model captures the number of key trust metrics based on centralities measurements and envisages the EFNN to classify the trusty and non-trust objects. The extensive experimentations are conducted using the real world datasets and various trust metrics were evaluated and compared with the other state-of-the-art trust learning models. The results demonstrate that the proposed model has outperformed with the other existing models by maintaining the accuracy of 95% to 94% with the decreasing untrusted rates and It illustrates conclusively that the LSTM's use of enhancing ensemble characteristics has shown to be more advantageous. While other models, like E-LSTM (90%) to 80%, Stacked LSTM (85% to 80%), SVM (85% to 79%), KNN (75% to 68%), and RF (65% to 58%), exhibit decreasing efficiency, the proposed approach illustrates that it is more productive as well as efficient for building an intelligent trust classification system that is appropriate for establishing trusted SIoT network communication.