Feature preprocessing algorithm based on one-class classifier

Aiming at the problem of features instability in specific emitter identification, this paper presents a based on one-class classifier feature preprocessing algorithm to detect the unstable features. The algorithm takes advantage of one-class classifier of property that can describe the distribution of given data sets. Its basic steps include: Firstly, we divide the feature time series into N segments with given length; secondly selecting one segment trains one-class classifier acquiring classification surface, and then remaining N-1 segments test the classifier and record the test accurate rates for every segments; finally we compare the minimize of test results with the setting threshold. If the minimize of test rate is lower than threshold in the test segments, we consider the feature is unstable. For a given set of data, experimental results show that the method is effective on determining the feature of stability.