An Interval Set Classification Based on Support Vector Machines

The input vector of standard support vector machine (SVM) is n-array attributes. Before new patterns are classified by trained SVM, the measurement of all attribute values is always necessary. In order to make incomplete information patterns can be classified correctly by trained SVM, we extend the inputs vector of SVM to interval input vectors where each unmeasured attribute of input is represented by an interval which includes its possible value, and the operation in classification function was extended to interval operation correspondingly. For the incomplete information input, the value of classification function is the interval operation result. When the output of classification function satisfies the classification condition, the incomplete information input pattern can be classified correctly. Meanwhile the attribute value prior knowledge about interval representation can be utilized fully in the proposed algorithm. Both theory analyze and experiments result all show the present algorithm is practical and effective, and the input attribute measurement cost can also be reduced