Securing Home IoT Network with Machine Learning Based Classifiers

Modern home network has traditional Ether-net/WiFi traffic along with emerging low power cross platform IoT traffic, which makes traditional Network Intrusion Detection Systems (IDS) approaches ineffective. This paper presents a deep neural network approach with a split architecture of Intrusion Detection System (IDS) specially suitable for home networks. The split architecture consists of multiple ML models and trained on two separate dataset for heterogeneous traffic. We also compare our model performance with reported different ML algorithms and found superiority of our model. The proposed model achieves 0.9694, 0.9625 and 0.9651 in precision, recall, and F1-score, respectively, for NSL-KDD dataset. Another interesting finding is that tree-based method and ensemble methods outperform our model in case the training dataset is unbalanced. An analysis of run-time implementation performance of the proposed IDS model is also discussed.