Enhancing liver disease diagnosis with hybrid SMOTE-ENN balanced machine learning models-an empirical analysis of Indian patient liver disease datasets.

Introduction: The liver is one of the vital organs of human body that performs some of the most crucial biological processes such as protein and biochemical synthesis, which is required for digestion and cleansing. A large number of patients are suffering from liver disease and hence it has become a life-threatening issue around the world. Annually, around 2 million people die because of liver disease, this accounts for around 4% of all deaths, due to factors like obesity, undiagnosed hepatitis, and excessive alcohol consumption. These factors accumulate and deteriorate the liver condition. Immediate action is necessary for timely diagnosis of the ailment before irreversible damage is done. Methods: The work aims to evaluate some of the traditional and prominent machine learning algorithms, namely, Logistic Regression, K-Nearest Neighbor, Support Vector Machine, Gaussian Naïve Bayes, Decision Tree, Random Forest, AdaBoost, Extreme Gradient Boosting, and Light GBM for diagnosing and predicting chronic liver disease. Also, real-world datasets often have imbalanced class distributions, causing classifiers to perform poorly, leading to low accuracy, precision, recall values and high misclassification. The Indian Patient Liver Disease (ILPD) datasets also face an imbalance issue. This work presents two hybrid models, namely SMOTEENN-KNN and SMOTEENN-AdaBoost, which can robustly handle the problem of imbalance in real-world datasets, in addition to improving the accuracy of liver disease prediction. We have also designed a hybrid model which involves the combination of Recursive Feature Elimination (RFE) for feature selection, SMOTE-ENN to tackle the problem of data imbalance and Ensemble learning for enhanced predictions. Results: The research work also proposed Hybrid Ensemble model on the ILPD and BUPA Liver Disorder Dataset. For the ILPD dataset, the model achieves an overall accuracy of 93.2% whereas for the BUPA dataset, the model attains an accuracy of 95.4%. The Brier Score loss for ILPD dataset is 0.032 and 0.031 for the BUPA Liver Disorder Dataset. Discussion: The research work highlights the potential of data balancing techniques and Ensemble models to improve predictive accuracy in liver disease diagnosis.