Designing an efficient PCA based data model for wireless sensor networks

Recent technology in wireless communication has enabled the development of low-cost sensor networks. Sensors at different locations can generate streaming data, which can be analyzed in real-time to identify events of interest. Wireless sensor networks (WSNs) usually have limited energy and transmission capacity, which cannot match the transmission of a large number of data collected by sensor nodes. So, it is necessary to perform in-network data aggregation in the WSN which is performed by aggregator node. Since, the nodes in WSN are vulnerable to malicious attackers and physical impairment; the data collected in WSNs may be unreliable. So, in this paper, we propose an efficient model based technique to detect the unreliable data. Data model is designed using the sound statistical multivariate technique called Principal Component Analysis (PCA). But as a drawback, it is not robust to outliers. Hence, if the input data is corrupted, an arbitrarily wrong representation is obtained. To overcome this problem, we propose two approaches namely Minimum Volume Ellipsoid (MVE) and Minimum Covariance Determinant (MCD) to design robust PCA which aids in design of a noise-free data model. The performance of proposed approach is evaluated and compared with previous approaches and found that our approach is effective and efficient.