Graded Classification of Liver Cirrhosis Using Machine Learning Algorithms on a Highly Unbalanced Dataset

Abstract

Liver cirrhosis is the fibrosis of liver caused by a long-term damage of the organ. This study classifies the disease in four classes based on a highly unbalanced dataset having 18 features and a data count of 6800 with labeled data of 465, 1507, 1322, and 3506 for the four classes using machine learning (ML) algorithms. Twelve ML algorithms have been deployed for the classification purpose which reflected the highest accuracy of 68.21% for the Histogram Gradient Boost Classifier. For further improvement of the accuracy, hyper-parameter tuning was done on all the ML algorithms which fetched the highest accuracy of 77.97% for the Gradient Boost Classifier (GBC). Further improvement of accuracy was observed with stacking model which furnished an accuracy of 84.24%. The stacked model comprised of the GBC as the meta-learner, and K-Nearest Neighbor (KNN), Xtreme Gradient Boost algorithm (XGB), Support Vector Machine (SVM), and the Light Gradient Boost Machine (LGBM) as the base-learners. To the best of our knowledge, this is the first attempt for graded classification of liver using all the ML algorithms, including hyper-parameter tuned and stacked models.