Anomaly Detection of IoT traffic Based on LSTM and Attention Mechanism

IoT traffic anomaly detection is a vital part of IoT security. However, it has the characteristics of long-time dimension and huge uneven feature information which brings big challenges. To address the above issues, we propose a new scheme for IoT traffic anomaly detection based on CNN-LSTM model. For data set pre-processing, we filter and zero pad the dataset, and then we offset and stitch it in the time dimension. It enables the CNN to extract features in the temporal dimension and reduces the computational complexity of the LSTM. We also catch the prediction result of each moment of the LSTM and output them to meet the timeliness required for IoT anti-intrusion. Moreover, the prediction results of the whole model guarantee the reliability of the scheme. Finally, we introduce the attention mechanism to the CNN-LSTM model, which improves the feature extraction ability of the model in the time dimension and solves the information overload problem. Validated by experiments on the IoT-23 dataset, the results show that the proposed model achieves an average accuracy of 91.6% which is about 6% increase over related algorithms, and demonstrates its effectiveness of the model in smart home application scenarios.