Wireless Sensor Networks Data Anomaly Detection: A Smart Approach

Wireless Sensor Networks (WSNs) are at the core of many critical civil and military applications where they are very amenable for deployment in some harsh/unsafe environments. The WSNs sensed data integrity and validity are very important in all applications, yet vital in some civil and military applications, where any faulty data might result in very serious consequences such as loss of human lives and precious commodities. In WSNs, sensors might report anomalous readings due to many factors related to their environment of deployment, energy level, computing capabilities, hardware failure, wireless media vulnerabilities, etc., which is a serious challenge. Hence, the need arises for immediate detection of such anomalies at the WSNs sink node, to guarantee the processing of accurately sensed data. In this paper, we devise a smart model to address a solution to the aforementioned problematic challenge. In our approach, we utilize different Machine Learning (ML) models in the process of anomaly detection in sensed data. The performances of the utilized ML models are compared, and the top performance model is incorporated into our smart anomaly detection model (SADM). In the design process of our new SADM, we experimented with over 2.2 million sensed data records from the Intel-MIT lab dataset to train and test many different ML models. The preliminary results of our SADM detection accuracy are very promising, which is very encouraging to proceed in this research direction, with different encoding data sets, to generalize the utilization of the new SADM model over a wider set of WSNs applications.