Global Outliers Detection in Wireless Sensor Networks: A Novel Approach\n Integrating Time-Series Analysis, Entropy, and Random Forest-based\n Classification

Wireless Sensor Networks (WSNs) have recently attracted greater attention\nworldwide due to their practicality in monitoring, communicating, and reporting\nspecific physical phenomena. The data collected by WSNs is often inaccurate as\na result of unavoidable environmental factors, which may include noise, signal\nweakness, or intrusion attacks depending on the specific situation. Sending\nhigh-noise data has negative effects not just on data accuracy and network\nreliability, but also regarding the decision-making processes in the base\nstation. Anomaly detection, or outlier detection, is the process of detecting\nnoisy data amidst the contexts thus described. The literature contains\nrelatively few noise detection techniques in the context of WSNs, particularly\nfor outlier-detection algorithms applying time series analysis, which considers\nthe effective neighbors to ensure a global-collaborative detection. Hence, the\nresearch presented in this paper is intended to design and implement a global\noutlier-detection approach, which allows us to find and select appropriate\nneighbors to ensure an adaptive collaborative detection based on time-series\nanalysis and entropy techniques. The proposed approach applies a random forest\nalgorithm for identifying the best results. To measure the effectiveness and\nefficiency of the proposed approach, a comprehensive and real scenario provided\nby the Intel Berkeley Research lab has been simulated. Noisy data have been\ninjected into the collected data randomly. The results obtained from the\nexperiment then conducted experimentation demonstrate that our approach can\ndetect anomalies with up to 99% accuracy.\n