머신러닝 기반 핸드볼 포지션별 경기력 평가체계 검증 : SHAP 중요도와 AHP 가중치 비교 = Validation of a Machine Learning-Based Performance Evaluation System for Handball Positions: A Comparison of SHAP Importance and AHP Weights

다국어 초록 (Multilingual Abstract)

This study aimed to validate a data-driven performance evaluation framework in women’s handball by classifying elite and non-elite players using machine learning techniques and comparing SHAP-based feature importance with expert-derived Analytic Hierarchy Process (AHP) weights. The dataset consisted of 11,861 match-level records of female field players who participated in six major international competitions, including the World Championships and the Olympic Games, from 2015 to 2024. Elite player classification was conducted using K-means clustering (K = 2). Subsequently, Random Forest and Extreme Gradient Boosting (XGBoost) models were developed to derive SHAP-based feature importance. Ensemble SHAP importance was calculated by normalizing and averaging the results from both models. The concordance between SHAP importance and AHP weights was examined using Spearman’s rank correlation coefficient and Kendall’s Tau. The results revealed position-specific key performance indicators: G/S was the most influential variable for center backs, pivots, and wings, whereas assists (AS) were most important for back players. The rank concordance between SHAP importance and AHP weights showed moderate agreement in some positions but was not statistically significant overall. These findings suggest that a data-driven evaluation framework can complement traditional expert-based assessment systems. This study provides empirical evidence supporting the objectivity and scientific advancement of performance evaluation in handball through the integration of machine learning and explainable artificial intelligence (XAI).