Optimal detection of attacks in wireless sensor networks using deep learning and Bayesian optimization

Abstract

Diagnosing attacks in wireless sensor networks (WSN) is a crucial concern in cyber security that impacts diagnosis models’ accuracy because of data imbalance and outliers existence. The present paper aims to design and perform a novel strategy for efficient attack diagnosis in WSN that applies DL models’ integration (DNN and CNN+LSTM) and Bayesian optimization. The present study checks concerns on data analysis such as various attack complexities and instances’ imbalance. Such concerns have been mentioned by applying techniques such as MinMax scaling, data balancing with ADASYN, polynomial feature engineering, and outlier elimination with DBSCAN. The outcomes of the experiments illustrate that the DNN model obtained an F1-Score of 85.71%, and appeared an accuracy of 99.14% which is an important development across conventional techniques. Such results illustrate that was presented technique could develop WSN security and possesses high capability for various attacks’ kinds’ diagnosis.