Malicious Node Detection Based on Self Encoder and Adaptive Kernel Density Estimation

Anomaly based malicious node detection is an important technology to protect the security of the Internet of things. In the existing research, statistics based detection usually builds probability density models based on existing samples However, with the increase of data dimension, it will be more difficult to build probability distribution model based on existing samples, and the calculation of distance density of data will become more complicated, which will lead to "dimension disaster". In order to solve the problem of difficult computation of high-dimensional features, this paper proposes a detection model based on self encoder and adaptive kernel density estimation. In this model, the high-dimensional features are compressed and reconstructed by self encoder, and the low-dimensional representation of the compressed potential space is combined with the reconstruction error as the low-dimensional representation of the high-dimensional features. Then, the model models the probability density function of low-dimensional features through adaptive kernel density estimation model and calculates the probability density as an anomaly score for detection. Experimental results show that the proposed model can achieve better detection performance on a variety of data sets.