Liver Disease Detection using Machine Learning: Impact of Feature Distribution, SMOTE, and Algorithmic Performance

Liver disease represents a serious medical problem which remains unnoticed until later stages because the earliest symptoms are minimal and difficult to identify. The research develops an artificial intelligence framework to enhance liver disease diagnosis through the analysis of structured medical records. The preprocessing of the dataset involved using Synthetic Minority Over-Sampling Technique (SMOTE) with variable imputation and encoding and scaling features and balancing classes. The core indicators that emerged from Exploratory Data Analysis (EDA) included age together with bilirubin levels along with albumin levels and protein levels. The Gradient Boosting and Extra Trees ensemble combined with Random Forest and XGBoost operated on an 80/20 split to perform performance analysis. Both XGBoost and Random Forest demonstrated reliable performance where XGBoost achieved 90.1% ROC-AUC and 77.6% accuracy while Random Forest had 81.3% accuracy but with 89.9% ROC-AUC. ROC-AUC measures revealed that Gradient Boosting and Extra Trees produced results comparable to each other when evaluated at 89.2% and 88.7% respectively. The research proves that combination models along with SMOTE augmentation and feature visualization systems demonstrate success in liver disease detection. The future work will integrate patient medical backgrounds and implement explainable AI systems and distributed learning models to preserve data privacy across various healthcare facilities thus enabling real-time diagnosis decisions in liver treatment settings.