An Improved Framework of Liver Disease Detection using SMOTE + ENN

Abstract

The diagnosis of liver disease at a preliminary stage is consequential for convalescent treatment. It is a demanding task for medical researchers to predict the early stages’ condition due to indistinct symptoms. More often than not, the symptoms become evident when it is too late. Furthermore, with the rise in the number of tests, the demands for faster and more accurate test systems have also risen. To overcome this issue, machine learning classification algorithms have been implemented on various datasets of liver patients. However, these datasets have a disproportionate number of cases for each class, making the model biased. We aim to solve this imbalance by applying a hybrid approach of SMOTE oversampling and Edited Nearest Neighbour undersampling techniques, which provide much cleaner data due to aggressive undersampling by  ENN.