Outlier detection based on data reduction in WSNs for water pipeline

Advances in data processing, electronics and wireless communications have made the vision of wireless sensor nodes an important reality. Wireless sensor nodes are cheap tiny sensor apparatus integrated with sensing, processing and short-range wireless communication abilities. Recent experimentations have been exploding in terms of usage and performance to improve the way of working in many contexts like the detection of outliers in a water pipeline. These pipelines are often subject to failure like erosion and sabotage that can cause high financial, environmental and health risks. Consequently, detecting damage and esteeming its location is very important. For this case, several techniques have been investigated in the research community. In this paper, we have constructed a novel leakage detection model based on Fisher Discriminant Analysis (FDA) and the Support Vector Machine (SVM) classifier for the detection of outliers based on Wireless Sensors Networks implemented in a water pipeline. Using this scheme, FDA is used to reduce the dimensionality of the pressure measurements and extract the optimal data for the classification process. Thus, SVM is used to perform the detection of the leaking pipe. The performance of our technique is evaluated in terms of accuracy and training time. As a result, the experimental measurements demonstrate that our method based on FDA coupled with SVM is the most efficient and accurate for detecting events in the context of water pipeline based on WSNs.