A weighted support vector machine classifier for contamination event detection in water distribuation systems

This paper presents an application of a weighted support vector machine (SVM) for the problem of contamination event detection in water distribution systems (WDS). The method utilizes general water quality measurements to construct a classifier for detecting abnormal behaviour, which is believed to imply an occurrence of a contamination event. The paper new contribution is the simultaneous analysis of the multivariate data in a high dimensional space, differing from the onedimensional parallel analysis that was conducted previously. Weighted SVM extend the method by considering that different input vectors make different contributions to the classifier. The resulted weights vector obtains two goals: blurring the difference between the sizes of the two training classes' data sets, and dealing with the time series attribute. A time decay factor yields higher importance to recent observations in the model. The classifier is updated constantly and exploits an increasing data base. The method was applied on a real WDS dataset and showed promising results.