Energy efficient outlier detection in WSNs based on temporal and attribute correlations

Support vector machines (SVM) have formulated the main concepts of machine learning, ever since their introduction. The one-class quarter sphere SVM has received recent interest, as it extends the concepts of machine learning to the domain of linear optimization problems with cost efficiency. This paper deals with the novel idea of a quarter-sphere SVM based only on temporal-attribute correlations. To avoid communication overhead the system complexity at individual sensor nodes is slightly increased. The outlier and event detection rate keeps up with the detection rate obtained via previous approaches with an added advantage of no communication cost.