A study on the effectiveness of machine learning models for hepatitis prediction.

Area of Science:

• Hepatology and Medical Informatics
• Machine Learning in Healthcare

Background:

• Hepatitis poses a significant global health burden with high mortality rates.
• Early prediction of hepatitis outcomes is crucial for effective management and improved public health.
• Current diagnostic and treatment advancements necessitate enhanced predictive capabilities.

Purpose of the Study:

• To apply advanced machine learning (ML) algorithms for predicting hepatitis outcomes.
• To identify key predictors of hepatitis survival using feature selection techniques.
• To compare the efficacy of various ML models in hepatitis outcome prediction.

Main Methods:

• Utilized the hepatitis dataset from the UCI repository (155 participants, 20 attributes).
• Employed the Boruta algorithm for feature selection, identifying Ascites, Varices, Bilirubin, Age, Spiders, and Alkaline Phosphate as key predictors.
• Evaluated logistic regression (LR) and six ML models: SVM, KNN, ANN, RF, AdaBoost, and XGBoost, using accuracy, precision, sensitivity, and F1 score.

Main Results:

• The Random Forest (RF) model achieved the highest accuracy (92.42%), precision (96.77%), and F1 score (96.00%).
• Key predictors identified by the Boruta algorithm include Ascites, Varices, Bilirubin, Age, Spiders, and Alkaline Phosphate.
• While RF showed high accuracy and sensitivity, its specificity was 33.33%; AdaBoost achieved the highest specificity (95.65%) but lowest sensitivity (50.00%).

Conclusions:

• The Random Forest model, combined with Boruta feature selection, is the most effective classifier for predicting hepatitis outcomes.
• ML-driven prediction of hepatitis survival can significantly enhance healthcare delivery and reduce the impact of communicable diseases.
• Findings support Sustainable Development Goal 3.3 by providing tools to combat epidemics like hepatitis.