Syncretic Feature Selection for Machine Learning-Aided Prognostics of Hepatitis

Despite recent advances in Machine Learning (ML)-based applications for clinical decision making, there is no objective method that can assist physicians in discriminating clinically significant features for predicting the prognosis of hepatitis. To this purpose, an expert-based objective feature engineering (FE) method is required to complement the physician’s subjective multi-criteria decision making and ML-based feature sets. Thus, this study proposes a novel feature selection (FS) method, which blends neighbourhood component analysis (NCA) and ReliefF by averaging their outcomes, coupled with a Lagrangian Support Vector Machine (LSVM) leveraged as an ML-based classifier for decision support to aid a binary classification of prognosis (survival or death) in patients with chronic hepatitis. Using this syncretic FS technique, which integrates the best performing FS methods tested, averaged feature ranks are obtained. Clinical data on 320 patients with hepatitis were obtained from two benchmark datasets from the University of California-Irvine database. The performance of the hybrid algorithms resulting from using statistical- and ML-based FE, both individually and in a syncretic manner, i.e., via a multi-expert ML system, was evaluated and compared. The proposed hybrid classifier NCA-ReliefF-LSVM, using an ML-based syncretic FS, led to the highest classification performance (AUC = 0.97/F1-score = 97.51, and AUC = 0.94/F1-score = 94.57) and the lowest computational cost (1 and 2 epochs, 13 and 11.67 s respectively) amongst all algorithms tested on both benchmark datasets. Thus, this study strongly supports the use of ML-based syncretic FS for predicting survival in individuals affected by hepatitis.