Machine Learning and Model Fusion for Multiclass Classification of Sarcopenia Severity = 사르코페니아 중증도 다중 클래스 분류를 위한 기계 학습 및 모델 융합

다국어 초록 (Multilingual Abstract)

This research introduces a machine learning approach for the multiclass
classification severity of sarcopenia, utilizing a model fusion framework. Sarcopenia
is defined by a gradual decrease in the mass, strength, and performance of skeletal
muscles, poses significant health risks for aging populations. While prior research
focused mainly on binary classification, this work aims to fill the gap by predicting
sarcopenia severity across four clinically meaningful stages: normal, risk (or mild),
impaired (or moderate), and severe. The proposed ensemble model integrates
classifiers— Gradient Boosting, Multilayer Perceptron (MLP), Random Forest —to
improve predictive accuracy. Feature selection is enhanced through dual-path
techniques, utilizing Random Forest and LASSO for linear and importance of
nonlinear features. Additionally, SHapley Additive exPlanations (SHAP) are
employed to ensure model interpretability, increasing clinical trust in predictions.
Performance evaluations show that the stacked model outperforms individual
classifiers, achieving an a macro F1 score of 0.9449, accuracy of 96.99%, and a
Cohen’s Kappa of 0.9738, with well-calibrated performance and generalizability.
This framework provides a reliable, understandable, and clinically applicable
instrument to categorize sarcopenia risk, offering potential for integration into clinical
decision support systems

목차 (Table of Contents)

• CHAPTER 1. Introduction 1
• 1.1. Research Background 1
• 1.2. Contributions 1
• 1.3. Motivation and Objectives 3
• 1.4. Thesis Organization3
• CHAPTER 2. Literature Review 5
• CHAPTER 3. Materials and Methods 10
• 3.1 Data Description and Study Design 11
• 3.2. Data Preprocessing 13
• 3.3. Model Development 14
• 3.4. Validation and Performance Metrics 16
• 3.5. Performance measurement 17
• CHAPTER 4. Results 19
• 4.1. Feature Selection 19
• 4.2. Binary Classification 20
• 4.3. Multiclass Classification 23
• 4.4. Overfitting Assessment 26
• 4.5. Model Interpretability 26
• CHAPTER 5. Discussion 28
• CHAPTER 6. Conclusion 32
• References 33