A novel hybrid algorithm for aiding prediction of prognosis in patients with hepatitis

Abstract

This study investigated the application of a novel hybrid artificial intelligence (AI)-based classifier for aiding prediction of the prognosis in patients with chronic hepatitis. Nineteen biomarkers on 155 patients with hepatitis from the University California Irvine Machine Learning repository were used as input data. Weights derived by applying the geometric margin maximisation criterion of a Lagrangian support vector machine (LSVM) were used for selecting the features associated with the highest relative importance towards the required classification, i.e. to predict whether a patient with hepatitis would have survived or died. Thus, the 19 initial features were reduced to the 16 most important prognostic factors and were fed into various AI-based classifiers. Results indicated an overall classification accuracy and area under the receiver operating characteristic curve of 100% for the proposed hybrid algorithm, the LSVM multilayer perceptron (MLP), thus demonstrating its potential for aiding prediction of prognosis in patients with hepatitis in a clinical setting.