This paper provides a comprehensive overview of Shapley value-based feature attribution methods in explainable artificial intelligence (XAI). Shapley value, rooted in cooperative game theory, has become a mainstream approach for explaining machine learning (ML) models due to its ability to provide interpretable insights into model predictions. The paper discusses the theoretical foundations of Shapley value, its desirable properties, and its practical applications in ML. A three-dimensional classification framework is proposed to categorize existing Shapley value-based feature attribution methods based on Shapley value type, feature replacement method, and approximation method. The paper emphasizes the application of Shapley value at different stages of ML model development, including pre-modeling, modeling, and post-modeling. It also summarizes the limitations of Shapley value and discusses potential directions for future research. The paper highlights the importance of Shapley value in improving model interpretability, fairness, and robustness, and explores its applications in feature selection, credit assignment in cooperative multi-agent reinforcement learning (MARL), data valuation, and model explanation. Despite its theoretical strengths, Shapley value faces challenges such as computational complexity, ambiguity in feature interactions, model sensitivity, and interpretability issues. The paper concludes that Shapley value remains a promising approach for feature attribution in ML, with potential for further development in model diagnosis, optimization, and integration of domain knowledge.This paper provides a comprehensive overview of Shapley value-based feature attribution methods in explainable artificial intelligence (XAI). Shapley value, rooted in cooperative game theory, has become a mainstream approach for explaining machine learning (ML) models due to its ability to provide interpretable insights into model predictions. The paper discusses the theoretical foundations of Shapley value, its desirable properties, and its practical applications in ML. A three-dimensional classification framework is proposed to categorize existing Shapley value-based feature attribution methods based on Shapley value type, feature replacement method, and approximation method. The paper emphasizes the application of Shapley value at different stages of ML model development, including pre-modeling, modeling, and post-modeling. It also summarizes the limitations of Shapley value and discusses potential directions for future research. The paper highlights the importance of Shapley value in improving model interpretability, fairness, and robustness, and explores its applications in feature selection, credit assignment in cooperative multi-agent reinforcement learning (MARL), data valuation, and model explanation. Despite its theoretical strengths, Shapley value faces challenges such as computational complexity, ambiguity in feature interactions, model sensitivity, and interpretability issues. The paper concludes that Shapley value remains a promising approach for feature attribution in ML, with potential for further development in model diagnosis, optimization, and integration of domain knowledge.