A Comprehensive approach to the detection of Liver disease using Machine Learning Techniques with comparison of different Oversampling Techniques on Imbalanced Liver Disease Dataset

Abstract

The liver is one of the most important organs in the body. It is responsible for controlling the chemical balance of the bloodstream as well as the removal of waste products among other vital functions. Liver disease is important to be diagnosed early on as symptoms do not begin to show until most of the liver is already damaged. Machine learning could be a crucial tool in the prediction of liver disease in patients which could lead to early diagnosis and also early treatment. In this study a dataset with 583 instances has been pre-processed and the imbalance had been handled in 5 separate ways, namely, Synthetic Minority Oversampling Technique (SMOTE), Adaptive Synthetic (ADASYN), Synthetic Minority Oversampling Technique and Conformal Clustering (CC), Synthetic Minority Oversampling Technique and Tomeklinks and Synthetic Minority Oversampling Technique and edited nearest neighbor (SMOTE+ENN). Then various machine learning algorithms like Decision Tree Classifier, Logistic Regression, Gaussian Naïve Bayes, Random Forest Classifier, K-Nearest Neighbors, and Support Vector Machine algorithms etc has been used. The experiment gave the best result when SMOTE+ENN was used as the imbalance handling technique with an accuracy of 98.37%. This accuracy was found using the support vector machine (SVM) approach. Therefore, this study shows the comparative analysis of the different imbalance handling techniques and the one which performs the best among each of these. It presents SMOTE+ENN as the best in case of this specific dataset.