Outlier Detection in Wireless Sensor Networks Using Machine Learning and Statistical Based Approaches

Outliers in wireless sensor networks (WSNs), stemming from harsh environmental conditions and limited processing and communication capacities of sensor nodes, pose a significant challenge to data reliability and quality collected by the network.Energyefficient outlier detection methods are crucial for prolonging network lifespan.This study introduces a two-phase approach to address this challenge.At sensor nodes, a lightweight statistical method based on mean and standard deviation detects and removes outliers, conserving energy.Early outlier filtering reduces data transmission, saving substantial energy due to the high energy cost of communication in WSNs.At the base station, several unsupervised Machine Learning algorithms, including One Class Support Vector Machine (OCSVM), Histogram Based Outlier Score (HBOS), Isolation Forest (IForest), K-Nearest Neighbor (KNN), and Cluster Based Local Outlier Factor (CBLOF), identify remaining outliers.The base station, with greater computational power and energy resources, can handle these tasks without the constraints faced by sensor nodes.Evaluation using realworld datasets demonstrates the effectiveness of our approach, achieving a 77.59% outlier removal rate at the node level while maintaining over 90% detection accuracy at the base station.By employing computationally light statistical methods at sensor nodes, reducing data transmission, and shifting complex tasks to the base station, our approach optimizes energy efficiency, minimizing consumption and reducing the need for frequent recalibration or maintenance, thereby extending the lifespan of the network.