Distributed SVM Classification with Redundant Data Removing

The biggest challenge faced by the distributed classification in wireless sensor networks (WSNs) is how to reduce the energy consumption in sensors for improving their service capacities. In this paper, an incremental Support Vector Machine (SVM) training method based on redundant data removing is proposed. Applying this method, distributed clustering is firstly performed on the data of sensors. Then boundaries are obtained to discriminate between clustered data and scattered data in clusters through Fisher Discriminant Ratio (FDR). The clustered data are regarded as redundant data and removed. Thus the number of data samples for training SVM is greatly reduced and then the computation is speed up in WSNs. Simulation results showed that the proposed method achieved goals of reducing energy consumption and keeping classification accuracies by decreasing time of training Support Vectors (SVs).