Enhancing Security in Smart Home Systems by Predicting Multiple Dos Attacks

Abstract

Internet of Things(IoT) is a network of embedded things that are able to sense,transfer data to other devices which can be used to automate mundane tasks. It is mostly sought after in Smart Buildings,energy sectors,Health care and even smart Vehicles. Smart Home systems are useful for Protecting Homes, Improving Ambience, and to automate regular home maintenance activities. It is also used to manage resources efficiently and also sought for taking care of elderly people for wellness assisted living. Despite the benefits, the recent attacks on CCTV Cameras, Baby monitors and other smart devices have clearly indicated the lack in Security and Privacy in Smart Home systems. Most of the smart home devices were transmitting data in plain text which can be eves dropped by Hackers. These devices were operating with default passwords which lets Hackers install malware piece of code by guessing the password combinations in a brute force method. The details or data thus leaked by these devices provide potential information about the whereabouts of the house inmates. To address these issues a Gateway device/Hub is introduced in Smart Home that monitors all the Smart Home devices connected to the Home Gateway. The Smart Monitor periodically scans the devices to check for open ports, and checks that the communication is happening only between registered known devices. To analyze the data with malware a bench mark data set IoT Network intrusion detection data set was analyzed and using different ensemble techniques like Random Forest, ADA Boost 9 Multiple Dos Attacks were classified. Ensemble Techniques were used to address the class Imbalance issue and Multiclass prediction was done using the data-set. XGBOOST algorithm showed more accuracy for different test data sets for binary as well as multi class classification. Anomaly detection based on Context of smart home usage scenarios is crucial to conclude whether an activity is an anomaly or not. If detected as Anomaly, to understand it a contextual based knowledge