SA-O2DCA: Seasonal Adapted Online Outlier Detection and Classification Approach for WSN

Abstract

Wireless Sensor Networks (WSNs) play a critical role in the Internet of Things by collecting information for real-world applications such as healthcare, agriculture, and smart cities. These networks consist of low-resource sensors that produce streaming data requiring online processing. However, since data outliers can occur, it’s important to identify and classify them as errors or events using outlier detection and classification techniques. In this paper, we propose a new and enhanced approach for online outlier detection and classification in WSNs. Our approach is titled SA-O²DCA for Seasonal Adapted Online Outlier Detection and Classification Approach. SA-O²DCA, combines the benefits of the K-means algorithm for clustering, the Iforest algorithm for outlier detection and the Newton interpolation to classify the outliers. We evaluate our approach against other works in literature using multivariate datasets. The simulation results, which encompass the assessment of our proposed approach using a combination of synthetic and real-life multivariate datasets, reveal that SA-O²DCA is stable with fewer training models number and outperforms other works on various metrics, including Detection Rate, False Alarm Rate, and Accuracy Rate. Furthermore, our enhanced approach is suitable for working with seasonal real-life applications as it can dynamically change the Training Model at the end of each season period.