Intrusion Detection Methods For WSN Based On Manifold Learning

As wireless sensor technology growing rapidly and applied in various fields. A wealth ofwireless sensor network (WSN) resources brought great convenience to people. But securityof wireless sensor network becomes a very important issue for it is vulnerable to a wide rangeof attacks due to deployment in the hostile environment and with limited resources.Detectinganomalous traffic is of primary interest in various WSN fields.Intrusion detection system isone of the major and efficient defensive methods against attacks in WSN. Internet detectionsystems almost use classifier algorithms to separate the anomalous network traffic data fromnormal. However, these single classifier systems fail to adapt to the WSN environment.WSNDesign of WSN needs to consider the energy of nodes firstly，then the efficacy ofalgorithm. Manifold learning in machine learning is a kind of dimensionality reductionalgorithm. It can map the samples from high dimensional space to low dimensional space, andfind out the intrinsic orderliness of the samples in high dimensional space. LaplacianEigenmap (LE) is a kind of robust and fast nolinear manifold learning algorithm, In this paper,we address the problem considering a method based on LE for detecting WSN networkanomalies of hierarchical model. This method use a kind of manifold learning-LE to reducethe features of anomalous traffic, reduce the number of data feature to be detected, thendecrease the data size corresponded between senor nodes and base station, node and node. Wewill use SVM (Support Vector Machine） to classify the decreased data and judge the kind ofthe data. Experiment results manifest that this method can find out the intrinsic manifold inthe high dimensional intrusion detection data, discover the rule in intrusion detection data,and decrease the energy consumption in the WSN, and it has the ideal detective efficiency.