Performance Evaluation of Machine Learning Techniques for DOS Detection in Wireless Sensor Network

Lama Alsulaiman,Saad Al-Ahmadi

doi:10.5121/ijnsa.2021.13202

Abstract

The nature of Wireless Sensor Networks (WSN) and the widespread of using WSN introduce many security threats and attacks. An effective Intrusion Detection System (IDS) should be used to detect attacks. Detecting such an attack is challenging, especially the detection of Denial of Service (DoS) attacks. Machine learning classification techniques have been used as an approach for DoS detection. This paper conducted an experiment using Waikato Environment for Knowledge Analysis (WEKA)to evaluate the efficiency of five machine learning algorithms for detecting flooding, grayhole, blackhole, and scheduling at DoS attacks in WSNs. The evaluation is based on a dataset, called WSN-DS. The results showed that the random forest classifier outperforms the other classifiers with an accuracy of 99.72%.

Highlights

Wireless Sensor Networks (WSN) is one of the important research topics in computer science
The WSN is a preferred solution for many applications in different fields such as medical and health care, telecommunications, WSN can be used during natural disasters to detect flooding, volcanoes, or earthquakes[1]
This section introduces preliminary information necessary for subsequent sections as well as some concepts related to Machine learning (ML) classification techniques and Denial of Service attack (DoS) attacks on WSNs

Summary

Introduction

Wireless Sensor Networks (WSN) is one of the important research topics in computer science. IDSs for WSNs are classified into four categories based on whether they employ signature-based, anomaly-based, specification-based, or hybrid technique[13].Signature-based techniquesare a common technique for detecting well-known attacks. This technique, which classifies traffic samples based on known patterns from the training dataset, is known for its high accuracy and low false-positive rate. While this method is effective, it suffers from some drawbacks; for instance, it is less efficient at detecting an unknown type of attack, since it needs to have a signature for the training dataset.

Objectives

Results

Conclusion