Classification of primary biliary cirrhosis using hybridization of dimensionality reduction and machine learning methods

摘要

The key functioning of human body depends on liver health. Liver performs numerous metabolic functions that also enable smooth working of other organs. Any form of illness in liver leads to liver diseases. These diseases are of many types out of which the most commonly occurring are hepatitis A, B, C, D and E, primary biliary cirrhosis (PBC), liver fibrosis, liver tumor, alcoholic liver disease, liver cirrhosis, fatty liver disease and autoimmune hepatitis. Presence of these diseases in various forms indicates the significance of accurate and timely diagnosis. This study accordingly aims to classify PBC stages using individual classifiers and hybrid models. Individual methods include linear discriminant analysis (LDA), diagonal linear discriminant analysis (DLDA), euclidean distance based k-nearest neighbors (KNN), and hybrid models include combination of LDA, DLDA and KNN with dimensionality reduction method. Simulations results showed that hybrid frameworks outperform individual classifiers in terms of classification performance. Furthermore, KNN based hybridization achieved a remarkable accuracy of 91.3%.