An Improved Unsupervised Anomaly Detection for Wireless Sensor Network using Machine Learning

Abstract

Wireless Sensor Networks (WSNs) are necessary and important network platforms for the future in line with the “Internet of Things” concept that arises recently. WSNs are used for tracking, monitoring or controlling in many applications for industry, habitat, health care, military, and other applications. However, the quality of data collected by sensor nodes is affected by anomalies occur due to various reasons, such as error readings, node failures, malicious attacks, and unusual events. Therefore, anomaly detection is necessary to ensure the quality of sensor data. Many challenges hinder the design of effective and efficient anomaly detection solutions for WSNs. These challenges include a high dimension of collected data especially in multivariate WSN applications, constrained resources, and dynamic streaming kind of sensor data, especially in environmental applications. The intent of this study is to create and develop an effective anomaly detection model for WSNs that efficiently utilizes the sensor ‘s limited resources and effectively detects anomalies in dynamic streaming WSN applications. To achieve this purpose an efficient Principal Component Analysis based Dimensionality Reduction scheme (PCA) is proposed based on Anomaly detection to minimize the number of features in multivariate WSN applications. The evaluation procedure employed find a comparison of the detection performance of the proposed anomaly detection model with the extant research was discussed as well. The study examined Unsupervised Machine Learning procedure for the labelled sensor data set of environmental signs and vital signs like (Humidity, Temperature, Blood presser, and Heart rate), and the results show that PCA (Principle Components Analysis) deemed the preferred anomaly detection algorithm for the present data set.