Open AccessA MACHINE LEARNING FRAMEWORK BASED ON VARIOUS NETWORK TRAFFIC CHARACTERISTICS TO IDENTIFY AND CLASSIFY THE DEFAULT BEHAVIOR OF IOT DEVICES ON A NETWORK
Author(s) -
Pagalla Bhavani Shankar,
Yogi Reddy Maramreddy,
P. Gayatri
Publication year - 2021
Publication title -
international journal of engineering applied science and technology
Language(s) - English
Resource type - Journals
ISSN - 2455-2143
DOI - 10.33564/ijeast.2021.v06i05.029
Subject(s) - computer science , internet of things , computer security , smart environment , artificial intelligence , data science , machine learning
The Internet of Things (IoT) is beingwell acquire to the next era of revolutionarygenerations amongst the new technologies. IoTtechnology being hailed so hard we had to stop inour society, smart homes, enterprises, and smartcities. Dynamics of smart one’s are increasinglybeing equipped with a profusion of IoT devices.Due to the tremendous upgradation of knowledgein various aspects impresarios of such smartenvironments may not even be fully aware oftheir working nature or principles of IoT devices,assets and functioning properly safe from cyberattacks. In this paper, we addressing thischallenge by developing a robust framework forIoT device classification using trafficcharacteristics obtained at the level of networklevel. As a part of robust framework, firstly, wehave a tendency to instrument a smartenvironment with 28 completely different IoTdevices, spanning cameras, lights, plugs, motionsensors, appliances and health-monitors. Wehave a tendency to collect and synthesize traffictraces from this framework infrastructure for aperiod of 6 months, a type of subset of which werelease as open data for the community to use.Second, we have to present or gifts the insightsinto the underlying network trafficcharacteristics using statistical and appliedmathematical attributes such as activity cycles,port numbers, signaling patterns and ciphersuites. Third, we have a tendency to develop amulti-stage machine learning based classificationalgorithm and demonstrate its ability to identifyspecific IoT devices with over 99% accuracybased on their network flow of activity. Finally,we have a tendency to discuss the trade-offsbetween cost, speed, and performance involved indeploying the classification network frameworkin real-time. Our study paves the way forimpresarios of smart environments to monitortheir IoT devices and assets for presence,functionality, and cyber-security withoutrequiring any specialized devices or protocols.