A Comparative Analysis of various Dimensionality Reduction Techniques on N-BaIoT Dataset for IoT Botnet Detection

Abstract

Internet of Things plays a vital role in our everyday lives in terms of economic, social, and commercial aspects. The widespread use of IoT devices has made them a prime target for cyber-attacks. IoT botnet attacks usually have a greater sensitivity to the consequences that might result from launching other attacks such as DDoS attacks and dissemination of sensitive information. For botnet detection in the IoT environment, deep learning mechanisms have emerged. But processing high-dimensional data is difficult, and it adversely affects DL-based botnet detection systems. Various dimensionality reduction methods have been proposed by researchers to address this concern. The purpose of this study is to examine and compare current mainstream dimensionality reduction methods. This will enable us to understand how reducing the number of features may lead to higher classification accuracy. Extensive tests are conducted on the NBaIoT dataset to verify the viability of PCA and auto encoder dimensionality reduction strategies. Results show that Auto encoder algorithm outperform PCA dimensionality reduction methods by the accuracy of 95.02%.