Liver Cirrhosis Classification using Extreme Gradient Boosting Classifier and Harris Hawk Optimization as Hyperparameter Tuning

This study proposes an early diagnosis model based on Machine Learning for liver cirrhosis classification using the Hepatitis C dataset, which is the leading cause of cirrhosis, from UCI ML. The classification is performed using the XGBoost algorithm because it provides high accuracy and time efficiency based on previous studies. However, these advantages depend on the combination of its hyperparameters set. XGBoost has a large number of hyperparameters, which can be time-consuming for researchers to manually configure. Therefore, this study proposes combining XGBoost with the Harris Hawks Optimization (HHO) algorithm for hyperparameter tuning. HHO is implemented with a hawk population of 40 and maximum iterations set at 25. The proposed XGBoost-HHO model provides an average performance of 99.34% for accuracy, MAR, MAP and 99.33% for Macro F1-score. These performances are achieved with the shortest processing time across 25 experiments compared to other combination models. The performance of the XGBoost-HHO model shows more significant increase in performance and reduction in overfitting compared to the standard XGBoost, SVM, RF models, as well as several other combined models including RF-HHO, SVM-HHO, XGBoost-PSO, and XGBoost-BA. Additionally, based on the feature importance analysis of the XGBoost-HHO algorithm, Alanine Aminotransferase (ALT), Protein, and Gamma-glutamyltransferase (GGT) contribute the most to the classification process, with gain values of 11.21, 9.51, and 7.98, respectively. Overall, the findings of this study show that the XGBoost-HHO algorithm combination provides competitive performance and can serve as an excellent alternative for liver cirrhosis classification in terms of both accuracy and time efficiency.