Efficient & Sustainable Intrusion Detection System Using Machine Learning & Deep Learning for IoT

Abstract

Everything is evolving toward IoT (Internet of Things) and online-based in our technological environment. The number of IoT devices and ubiquitous computing systems are growing exponentially. This also increases the risk of network breach. To cater this issue many researchers proposed different techniques and get great results but it can be better since everything in online and it's a matter of security and privacy. This paper presents an efficient and sustainable intrusion detection system by the concatenation of two well-known state of the art “kitsune” datasets (ARP MITM and SSDP Flood). Random Forest, decision tree, and Bi-LSTM (Bi-Directional Long Short Term Memory) were implemented in different training and testing ratios and different numbers of layers. Performance measures show that all the models achieved over 99% accuracy but random forest outperforms both models on the concatenated dataset. Both attacks are determined by the given model hence increasing the performance and the system will notify in case of any malicious activity.