Evaluation of Naïve Bayesian Algorithms for Cyber-Attacks Detection in Wireless Sensor Networks

Abstract

Wireless Sensor Network (WSN) is one of the Internet of Things (IoT) operating platforms, which has proliferated into a wide range of applications. These networks comprise many resource-restricted sensors in terms of sensing, communication, storage, and power. Security becomes a critical concern to protect the network of scarce resources from any malicious activities that target the network. Several solutions have been presented in the literature; however, machine learning has proven its appropriateness in designing energy-efficient detection measures for cyber-attacks targeting WSNs. This paper presents a WSN security performance evaluation of three Naïve Bayesian machine learning classification technique variants: Gaussian Naïve Bayes, Multinomial Naïve Bayes, and Bernoulli Naïve Bayes, compared to three well-known base algorithms: K-Nearest Neighbors, Support Vector Machine, and Multilayer Perceptron. We applied Spearman correlation as a univariate feature selection. The specialized dataset, WSN-DS, was used for training and testing purposes. The performance of the six classifiers was evaluated in terms of accuracy, probability of detection, positive prediction value, probability of false alarm, probability of misdetection, memory usage, processing time, prediction time, and complexity.