Machine Learning Approach For Detection Of Attacks In Wireless Sensor Networks

Abstract

The widespread use of Wireless Sensor Networks (WSNs) in several areas with special newlinefocus on crucial data collection necessitates to maintain the highest level of security in newlineWSNs. Several security threats and vulnerabilities pose risks to the seamless operation newlineof WSNs. Establishing a resilient system to counteract these security challenges is vital newlinefor the proper functioning of WSNs. newlineA number of attack detection techniques in WSNs are available in the current literature. newlineHowever, with the emergence of Machine Learning (ML) / Deep Learning (DL) newlinetechniques, there exists scopes for developing attack detection techniques using ML/DL newlinetechniques that can deliver better performance. The thesis focuses on development of newlineML/DL-based attack detection techniques in WSNs. The use of supervised learning, newlineensemble learning, and unsupervised learning approaches are considered. newlineIn this thesis, various ML/DL-based approaches for the detection of attacks in WSNs newlineare proposed and presented. The first proposed method uses CNN-based attack newlinedetection technique and uses median, mean, STD, skewness and kurtosis as feature newlineselection methods. The second proposed method uses CNN-based attack detection newlinetechnique and uses mutual_info_classif and f_classif functions as feature selection newlinemethods. The third proposed method is based on DCNN and uses a distinct optimization newlinestrategy to fine-tune the filter sizes, number of filters in a fully connected layers, along newlinewith the input vector. Fourthly, a cascadeand#8209;structured deep MLP-based ensemble newlinelearning and averaging ensemble learning-based approaches for detection of attacks in newlineWSNs has been proposed and presented. The fifth proposed method is based on newlineunsupervised learning that uses FCM (Fuzzy-c-means) and K-means along with PCA newline(Principal Component Analysis) as a dimensionality reduction. Finally, an optimized newlineCNN-based attack detection that employs PSO (Particle Swarm Optimization) newlinealgorithm to optimize the hyperparameters of the CNN model has been proposed and newlinepresented. newline