Semi-Supervised Learning Based Trust Evaluation for Underwater Wireless Sensor Networks

In recent years, trust mechanism has gradually become an effective scheme to deal with the internal attacks of underwater wireless sensor networks (UWSN). However, most of the existing trust models are based on traditional machine learning algorithms, which require a large amount of data training to improve the accuracy of the model. Therefore, these models still face the challenge of insufficient data in UWSN. In this paper, we propose a trust evaluation method based on Semi-Supervised learning (TESS). We consider the difficulty of underwater data collection and the lack of valid data. TESS uses a Semi-Supervised classification method based on Generative Adversarial Networks (GAN) to classify the collected trust parameters. This method can train high-precision detection models using a small amount of labeled data and a large amount of unlabeled data. Simulation results show that compared with LTrust and STMS, the accuracy of TESS under Bad-mouthing attacks is respectively improved by 26.45% and 26.78%.