Improved Classification of Liver Disorder and Blood Transfusion Donor Data Using Mixed Kernel SVM

Accurate detection and classification of various ailments has been a critical issue in the field of medical science and research since a long time. There are many mathematical, statistical and machine learning approaches exist for this purpose which results in limited success due to various reasons. This paper proposes an approach consisting of mixed kernel based support vector machine along with tuning of appropriate kernel parameters to improve classification performance. The mixed kernel based approach uses an iteratively determined optimal combination of different standard kernels, to produce better classification accuracy. Tuning of kernel parameters like polynomial degree in case of polynomial kernel and kernel width in case of Radial basis function kernel, is done iteratively to get best possible results. Experiments with two benchmark datasets, 1. BUPA liver disorders dataset and 2. Blood transfusion service center dataset, demonstrate the improvement in classification performance.