A Novel Approach for Outlier Detection Using Two-Phase SVM Classifiers with Cross-Training Method for Multi-Disease Diagnosis

Abstract

Data mining can be effectively used for disease diagnosis but diagnosis of complex-linked diseases, such as TB and HIV, is critical task. Outlier detection can be used on health datasets for identifying rare and abnormal instances for diagnosis purpose. Support vector machine (SVM)-based outlier detection can be used for detecting Outliers from the health dataset and it can be possible in two ways: two-class SVM and one-class SVM. But both classifiers suffer with non-availability of accurate class labels and long training time. So, the main objective of this work is to identify abnormal instances from the health datasets for multiple disease diagnosis combining both the SVM classifiers and introduced cross-training method for solving issues of both classifiers. By comparing the proposed approach with other existing classifiers such as two-class SVM and one-class SVM, the proposed classifier detects almost exact number of outliers from both the datasets with a good accuracy and less training time.