Low-complexity Anomaly Detection Method based on Feature Importance using Shapley Value

The increasing popularity of Internet of Things (IoT) devices has brought significant security challenges to IoT networks. However, most deep learning-based anomaly detection solutions often require high computation performance so that it is difficult to be implanted on low-end IoT devices with limited power and memory capacity. In this paper, we propose a low-complexity network anomaly detection method based on feature selection using the Shapley value for the Isolation Forest algorithm. The proposed feature selection method using the Shapley value can reduce the dimension of input data, thereby improving the performance with reduced computational complexity. We provide simulation results to demonstrate the effectiveness of the proposed method. The results show that the proposed method based on Isolation Forest achieves comparable performance to the deep learning method based on neural networks while using fewer dimensions than the deep learning method.