Detecting Anomalies in IoT Sensor Data with Robust Principal Component Analysis (RPCA)

This work studies the use of RPCA in IoT sensor data for anomaly detection. However, due to the fact that IoT data is complex, high-dimensional, noisy, and frequently has a dynamic pattern, it is hard to employ regular anomaly detection techniques. RPCA, however, gives a quite potential solution as it decomposes the data into low-rank and sparse matrices, in this way the outlying points can be separated for isolation. The effectiveness of the proposed framework is being validated, compared to the state-of-the-art schemes for IoT, and yields high detection accuracy and low FPR/FNR rates in real-world IoT datasets with noise and different severity of anomalies included. It should also noted that the computational complexity of RPCA is bigger than of traditional methods but it offers much better results. The present work can be considered as a valuable contribution to the field of IoT due to a highly reliable and scalable method that can be used for anomaly detection and recognition of IoT systems vulnerabilities.