Global Sensitivity Analysis of Binary Response Model Based on Shapley Value

The issue of the binary response model finds extensive applications in engineering practice, such as experimental design, medical diagnosis, computer science, and other related fields. Global Sensitivity Analysis (GSA) represents a crucial research direction. However, challenges faced by binary response model problems include limited data availability, nonlinearity of the model, and factor correlation. Traditional methods of global sensitivity analysis (GSA) are constrained to small sample sizes and fail to account for the intricate interplay among factors. To overcome these limitations, this paper proposes a novel approach for sensitivity calculation by integrating the Shapley value method from game theory and economics. Through comparing factor rankings obtained from conventional GSA methods, exploring the distribution relationship between the sensitivity index and individual sample point values, and considering real-world experiences and underlying mechanisms, we demonstrate that our proposed sensitivity analysis method not only addresses challenges related to data scarcity and model nonlinearity in binary response models but also effectively handles factor correlations.