Multi-Stage Intrusion Detection System for IoT Network using Deep learning

Abstract

With the advent of IoT, an advanced era of communication has been developed as everything around us could be connected to a network. The last decade has seen a growing trend towards developing and deploying a large scale of IoT devices. In general, a typical IoT network will exchange an enormous amount of data every second, thus, these devices are prone to security threats. An intrusion De tection System (IDS) is one of the most common security solutions for identifying cyber-malicious attacks and threats. However, the main challenge faced by many IDSs is the endless increase of new threats that current systems do not recognize. This project is divided into two parts. In the first part, we aim to introduce multi-stage and efficient intrusion detection system in IoT networks using supervised machine learning (ML). The system pro posed will be able to learn the discriminative feature representation automatically from a large amount of data and then accurately clas sifies the different classes efficiently. In order to achieve this project, we build a model to see whether or not the traffic encountered in the network is normal. Then, the system proposed will detect the different attack classes and sub-classes at different stages. Stage 1 xii List of Tables classifies the network packet as normal or anomaly, stage 2 catego rizes the category of attack, and stage 3 classifies the sub-category. We plan to validate the robustness and effectiveness of the system proposed using well-known IoT benchmark datasets. After that, our model will be evaluated using different performance metrics and compared with the state-of-the-art techniques to demonstrate its im provements over other related systems. In the second part, transfer learning is used to improve the perfor mance of the target domain model. To increase the availability of our prior knowledge about the target future data, real-world applica tions can potentially integrate related data from a different domain in addition to data in the target domain. By transferring valuable information from data in a related domain for use in the target activ ities, transfer learning solves such cross-domain learning difficulties. The tabular data is converted into images using DeepInsight. Then CNN model is trained on dataset 1. This model is then used as a base model and further trained on sparse dataset 2 in order to de termine the knowledge (weights) learned from dataset 1 assist target model to perform better on the target data set.