Abstract:

SVM (Support Vector Machine) is a useful technique for data classification. Normalize data before applying SVM is very important. For lack of better prior information, it is common to normalize attributes to the same range with the same method. Three of the most useful method to normalize attributes are: mean 0 and standard deviation 1, midrange 0 and range 2 or when the significance of magnitudes is non-linear the attribute values can be scaled by taking logarithms (or by taking cube roots) then transforming to midrange 0 and range 2. In this page we propose a method to use GA (Genetic Algorithm) to select a normalize method for each attribute. Our experimental results show that the mothod we proposed better than the method is often used that normalize attributes by same method