Detection of Abnormal Blood Samples Using Support Vector Machines with Radial Basis Function
Qiu De · Mini-micro Systems · 2003
This paper describes an approach for the detection of abnormal blood samples using Support Vector Machines(SVM). The problem of abnormality detection falls in more general category of non linear binary classification, but it is with the remarkable property that the training samples are much imbalanced and the margins of the classifying boundary to each class are expected to be unequal. Considering the particularity, we suppose some abnormal blood samples on a hypersphere. These supposed abnormal blood training samples, together with the practical normal blood training samples, are mapped into a high dimensional inner product space by Gaussian Radial Basis Function(RBF) kernel, where they can be separated by a linear hyperplane. Through adapting the width of RBF and the margin of separating hyperplane, the boundary that surrounds the subspace of normal blood samples closely and is with the best result of abnormality detection can be determined. This approach achieved the good result that the false alarm rate is 3.19%, the missing alarm rate is 6.38% and the accuracy rate is 90.43%.