A New IDS for Smart Home based on Machine Learning

IoT environments are highly diverse with regard to devices, applications, and communications protocols. Consequently, they are highly vulnerable to many new attacks specific to this particular network. Existing Intrusion detection systems have shown their inefficiency in loT and this is for many reasons related basically to the limited computation capability of the devices, their mobility, the inherent Internet connectivity, and the large scale of the IoT network. Thus, a lightweight and efficient intrusion detection system designed for loT is required. Inspired by the success of Machine Learning in many fields and its potential in attack detection, we propose in this paper an intrusion detection system for Smart Home. The main goal is to design a model that detects different attacks on different Smart Home devices. We aim to enhance the detection capabilities of our IDS by exploring different features allowing us to classify the input traffic as benign or malicious. More importantly, we seek to develop specific models to predict the type of attacks per device. To this end, we trained two ML and two DL models using an loT dataset named 10T/IIoT. The obtained results show that the ML algorithms trained after applying a feature selection technique outperforms the model where all features are used in the training. Besides, the ML models allow reaching accuracy values competing with the accuracy of DL models, further, they outperform DL models for some devices/attacks.