非酒精性脂肪肝疾病（Non-Alcoholic Fatty Liver Disease, NAFLD）為全球最常見之慢性肝病，隨著肥胖、代謝症候群盛行率逐年上升，非酒精性脂肪肝病NAFLD 已成為重大全球公共衛生議題。其疾病進程範圍涵蓋單純性脂肪堆積至非酒精性脂肪性肝炎，最終可能導致肝纖維化、肝硬化甚至肝細胞癌。因此，發展一套高效、準確且低成本之非侵入性篩檢工具，對於早期發現及治療 NAFLD 具有重要意義。
本研究旨在建構一套 NAFLD 多元分類模型，研究資料來源包含多項臨床指標，經由資料前處理及特徵篩選後，選用隨機森林、XGBoost、SVM及邏輯斯迴歸作為基礎分類器，並進一步結合投票法（Voting）與堆疊法（Stacking）進行集成學習模型設計與效能評估。實驗結果顯示，集成模型於受試者工作特徵曲線曲線下面積（AUC-ROC）、準確率、精確度、召回率與F1分數等多項分類評估指標皆優於單一機器學習模型，特別是堆疊模型展現出最優異的分類能力。研究成果有助於臨床醫學透過非侵入性臨床數據進行疾病分類，提升大規模健康篩檢效率，並作為臨床決策支援之輔助工具。整體而言，本研究所建構之集成分類模型，展現出於脂肪肝疾病分類任務中的應用潛力，為未來智慧醫療發展提供有力參考。

Non-alcoholic fatty liver disease (NAFLD) is the most common chronic liver disease globally and a growing public health issue driven by increasing rates of obesity and metabolic syndrome. The disease spectrum ranges from simple hepatic steatosis to non-alcoholic steatohepatitis (NASH), which may further progress to liver fibrosis, cirrhosis, and even hepatocellular carcinoma. Therefore, developing an efficient, accurate, and cost-effective non-invasive screening tool is crucial for the early detection and management of NAFLD.
This study developed a multi-class classification model for NAFLD severity using clinical data. After preprocessing and feature selection, Random Forest, XGBoost, SVM, and Logistic Regression were employed as base classifiers. Ensemble learning methods, including Voting and Stacking, were applied to enhance predictive performance. Results demonstrated that ensemble models, particularly stacking, outperformed individual classifiers across key metrics (AUC-ROC, Accuracy, Precision, Recall, F1-score).The proposed framework offers a cost-effective, non-invasive tool for NAFLD classification, supporting large-scale screening and clinical decision-making. This work highlights the potential of ensemble learning in advancing precision healthcare for liver disease management.