Sign in Subscribe
Artificial intelligence

Explainable Artificial Intelligence: A Comprehensive Survey

Strategic Insights

— 1 min read

Research conducted at St. Petersburg State University has described and investigated the Explainable Artificial Intelligence (XAI) methods based on the Shapley value, one of the most well-known imputation concepts in cooperative game theory. The study aimed to identify the importance of features when building statistical and machine learning models, analyzing the quality of explanation results given by different XAI methods, and their running time. The research concluded with recommendations for choosing an appropriate XAI method depending on the type of model to be explained, required running time, and the quality of the explanation results.

Key Takeaways:

  • The study investigated the Explainable Artificial Intelligence (XAI) methods based on the Shapley value, a concept in cooperative game theory used to estimate feature importance.
  • Five datasets were used for experiments, two of which were high-dimensional with over 100 features, to analyze the quality of explanation results given by different XAI methods and their running time.
  • The research concluded that recommendations should be given for choosing an appropriate XAI method depending on the type of model to be explained, required running time, and the quality of the explanation results.
  • Feature Importance (FI), Permutation Feature Importance, and LIME methods, which do not rely on game-theoretic ideas, were compared with Shapley value-based methods: SHAP, KernelSHAP, SamplingSHAP, FastSHAP, and ShapG.
  • Authors Elena Parilina, Chi Zhao, and Jing Liu contributed to the research, with Elena Parilina available for further information at St. Petersburg State University.

Statistics:

  • Five datasets were used for experiments, including two high-dimensional datasets with over 100 features.
  • Experiments were conducted on high-dimensional datasets with 100+ features.
  • The study analyzed the quality of explanation results given by different XAI methods and their running time.

Sources:

  • "The Shapley Value Contribution To Explainable Artificial Intelligence: a Comprehensive Survey." Dynamic Games and Applications, 2025.
  • NewsRx. "Researchers from St Petersburg State University Provide Details of New Studies and Findings in the Area of Artificial Intelligence (The Shapley Value Contribution To Explainable Artificial Intelligence: a Comprehensive Survey)." Journal of Engineering, October 27, 2025.

Read more