Hybrid Intrusion Detection Method Based on K-Means and CNN for Smart Home

This paper reviews the problem of instrusion detection for Smart Home and different approach to detect instrusion. A hybrid instrusion detection method based on Convolutional Neural Networks(CNN)and K-means is proposed in this paper. At smart home device node, K-means is used to generate the rule base by clustering, then Principal Component Analysis(PCA)is used to extract the dimensionality reduced features. During the test process, PCA is also used to extract the dimensionality reduced features, the feature matching is performed with the rule base to determine the intrusion data. At the smart home server side, a CNN model is proposed to detect the specific type of intrusion. Combined with Synthetic Minority Oversampling Technique(SMOTE)and undersampling techniques, the CNN model has great performance in reducing missing report rate(MRR)in minority categories. The results of the experiment conducted in KDD99 dataset show that such a hybrid method can improve the detection rate of smart home intrusion detection system and reduce MRR in minority categories.