Advanced Machine Learning Approaches for Liver Disease Prediction: A Focus on Feature Selection and Hybrid Ensemble Model

This paper classifies liver disease using advanced techniques of authentication such as comprehensive datasets upon which the advance machine learning methods are based. In the methodology, careful data preprocessing included data cleaning and treatment of class imbalances through Random Oversampling, outlier detection by the Local Outlier Factor method, and Min-Max Scaling for feature standardization to ensure uniformity in the presence of such features across the overall dataset. However, in the case of LARS feature selection for LASSO-ANOVA Recursive Selection approach, the hybrid combination of LASSO algorithm and Recursive Feature Elimination and ANOVA-a that successfully picked five major features is involved, namely, Direct Bilirubin, Aspartate Aminotransferase, Alkaline Phosphatase, Age, and Albumin. These features enhance quite a margin above the predictability of the model. During the model-building phase, an ensemble fusion strategy was employed using Random Forest, AdaBoost, and CatBoost algorithms that utilize the overall strength of the algorithm. Outstanding performances were delivered with 96.67% accuracy in the test dataset, a weighted F1 score of 90.62%, precision of 90.67%, and a root mean square error of 0.3055. This scientific work brings to the fore the potential application of advanced algorithms of machine learning in the diagnosis of liver disease, providing health professionals with information crucial to enrich their clinical decision-making and the ultimate outcome for patients suffering from these diseases.