The Application Of Support Vector Machine And Non-negative Matrix Factorization Theory Method

The vector support machine (SVM) technology is a new model which is proposed by V. Vapnik in the mid 90s, it is based on the statistical learning theory (SLT) which is different from the traditional classification methods. It is a new data mining tool using optimization method to deal with the machine learning problem. Recently years with the deep of the research, both its theory and actual application achieve a great breakthrough. It appears a greater advantage to the dimension reducing and over fitting problems, also it solves regression and pattern recognition problem successfully. In recently years, SVM has been a hot topic in the machine learning filed.The non-negative matrix factorization(NMF)technology firstly proposed by D.D.Lee and H.S.Seung in 1999 in《Nature》,it is able to learn parts of the image that is psychological and physiological evidence for parts based representations in the brain. Non-negative matrix factorization distinguishes from the other methods because its use of non-negativity constrains. The constrains lead to a parts-based representation because they allow only additive, not subtractive, combinations. NMF has cause huge attention at the very beginning, it is applied in face recognition first and appears a good result, recently years it has involved in a lot of areas such as signal processing, biomedical engineering, pattern recognition, network security and so on.Around with SVM and NMF, there are two main aspects in this paper as follows:1. It first utilizes normal distribution based on the probability of membership functionπto calculate the ambiguity, due to the normal distribution feature, take the distribution of the data into consideration to calculate the membership. The outputs can reflect the characteristic of the membership more accurately and improves the classification accuracy.2. There is a new method which integrates NMF and non-linear dimensionality reduction Isomap is proposed. The global dimensionality reduction method can discover the inner structure and relativity of the data, it makes the high dimension data visualization on lower space .The NMF is a local data mining method which can extract the local feature of a image, it can describe the relevant image distribution on the space of base matrix In image retrieval experiment, the method can obtain information more precise and improve the accuracy.