Neighboring Information Exploitation for Anomaly Detection in Intelligent IoT

The identification of anomalies has become increasingly important for the security of sensory data gathering in the intelligent Internet of Things (iIoT). The current approaches might not be applied to the general cases of anomalies, i.e., both long-term and short-term anomalies, as well as not be suitable with real-time applications such as natural disaster monitoring and early warning systems. To address this challenge, this paper proposes a comprehensive approach, named DWT-PCA Anomaly Detection (DAD) to detect both long- and short-term anomalies. DAD bases on the combination of Discrete Wavelet Transform (DWT) and Principal Component Analysis (PCA) to improve the system performance. In particular, we first utilize the DWT to extract approximation coefficients and detail coefficients from the input data which are capable to highlight long-term and short-term anomalies, respectively. We then exploit the spatial-temporal correlation of neighboring sensors by applying PCA on these coefficients to obtain a high detection accuracy. Numerical experiments based on the real dataset of Intel Berkeley Research reveal that the proposed scheme obtains higher accuracy and a lower false-positive rate on detecting three typical anomalies: drift, noise, and outlier, comparing to existing schemes.