Machine Learning-Based Network Intrusion Detection System for IoT Environment

Presently, the world has so much advanced technology which is advancing every day. With the help of advancement and evolution in technology, the world has a huge number of heterogeneous devices which are connected through the Internet. The number of devices connected through the Internet is also increasing day by day. Each device shares information, individuals share data with another individual and a large number of people use applications to share the information and in businesses, data exchange is happening all around the world. In a single day, there are millions of information exchanges happening but with the advancement in technology there is an increase in threat in the security of the information exchange. With the increase in the number of connected devices, there is also an increasing security threat. To fight against malicious behavior and threats as a defense line in communication networks. Network intrusion detection systems (NIDSs) are extensively utilized intrusion detection systems, but there is no standard methodology for the comparison of different NIDS. Several proposed papers do not mention the important steps for the validity of NIDS which makes comparison difficult, if not impossible. The paper focuses on the processes that must be followed in order to make a valid comparison and evolution for a NIDS. The paper performed a stepwise methodology to the UGR’16 dataset to address the network attack detection problem. In this paper, different machine learning techniques are applied to find the network threat and select the best performing machine learning technique for detection of threat, so as to increase in performance of the detection of security threats. In our research, a very systematic experimentation is done in various steps. Firstly, processed and modified the dataset with feature engineering followed by feature selection. Further on, data preprocessing followed by hyper parameter tuning. After our data was processed and modified by previous steps, we finally used it to train our machine learning models. The efficiency of the proposed model came out to be better than other models based on similar approaches.