KNN-SVM Classifiers in Complex Diagnosis

Abstract

Abstract In many applications, classification plays an indispensable role due to its powerful detection and diagnosis function. Especially in real data on disease, the detection of important factors and the diagnosis of the result usually bring huge contributions to patients. Simultaneously, complex problems in real data such as imbalanced data and missing data also lead to more challenges and difficulties. The ignorance of missing data will undermine study efficiency, and sometimes introduce substantial bias. Imbalanced data tends to be overwhelmed by the majority classes and ignoring the minority ones. The paper develops new support vector machine classifiers using k-nearest neighbors’ information (KNN-SVM), to impute missing data by calculating k-nearest neighbors’ statistical characteristic values and to interpolate some new samples between k-nearest minority class examples. As comparisons, the paper uses different kernel functions in KNN-SVM classifiers to show the different performances in disease diagnosis accuracy.