A Generic Framework for Soft Subspace Pattern Recognition
Dat Thanh Tran, Wanli Ma, Dharmendra Kumar Sharma, Len Tien Bui, Trung Le · University of Canberra Research Portal · 2009
We have proposed a generic framework for soft subspace pattern recognition. The framework has been designed for continuous hidden Markov model. The framework for fuzzy subspace Gaussian mixture model has been extracted by setting the number of states in continuous hidden Markov model to one. With an assumption on covariance matrix and density, a fuzzy subspace model for vector quantization has been determined. The proposed methods are based on fuzzy c-means modeling to assign fuzzy weight values to features depending on which subspace they belong to. We have also applied the vector quantization model to anomaly network detection problem. We have used the KDD CUP 1999 dataset as the sample data to evaluate the proposed methods. The fuzzy subspace vector quantization method outperformed the standard vector quantization model.