A machine learning technique in a multi-agent framework for online outliers detection in Wireless Sensor Networks

Wireless Sensor Networks enable flexibility, low operational and maintenance costs, as well as scalability in a variety of scenarios. However, in the context of industrial monitoring scenarios the use of Wireless Sensor Networks can compromise the system´s performance due to several factors, being one of them the presence of outliers in raw data. In order to improve the overall system´s resilience, this paper proposes a distributed hierarchical multi-agent architecture where each agent is responsible for a specific task. This paper deals with online detection and accommodation of outliers in non-stationary time-series by appealing to a machine learning technique. The methodology is based on a Least Squares Support Vector Machine along with a sliding window-based learning algorithm. A modification to this method is considered to improve its performance in transient raw data collected from transmitters over a Wireless Sensor Networks (WSNs). An empirical study based on laboratory test-bed show the feasibility and relevance of incorporating the proposed methodology in the context of monitoring systems over Wireless Sensor Networks.