Data Reduction Using NMF for Outlier Detection Method in Wireless Sensor Networks

Abstract

Nowadays, in wireless sensor networks (WSNs) field, the advances in electronics, wireless communications, and data processing have become an important reality. Wireless sensor networks are employed to eliminate problems occurred in health care, monitoring, agriculture, etc. Data reduction is considered as the best method that reduces dimensionality of the database. So, it helps outlier detection technique to classify data during training. In our work, we have constructed a newest data reduction process using non-negative matrix factorization (NMF). This method finds out the nature of data it is regular or outlier. Also, it can give a highest performance. Compared to various methods like Fisher Discriminant Analysis (FDA) and Principal Component Analysis (PCA), our method based NMF are considered as the most efficient and accurate for detecting outlier in WSNs. As real datasets, we use LUCE, Intel Berkeley and Grand St-Bernard. So based on the obtained results, our method is considered as a perfect one used in wireless sensor networks.