Distributionally Robust Shapley Value Estimator: An Application to Industrial Predictive Modeling
作者单位：College of Metrology Measurement and Instrument China Jiliang University School of Mathematics Hangzhou Normal University School of Information Science and Engineering Hangzhou Normal University School of Statistics Jiangxi University of Finance and Economics
会议名称：《第35届中国过程控制会议》
会议日期：2024年
学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）]
关 键 词：Explainable artificial intelligence deep learning Shapley value predict modeling
摘 要：Predictive models based on deep learning techniques have become one of the most popular data-driven approaches in the industrial process. Nevertheless, these models often lack interpretability due to their black-box nature. In order to interpret the predictions of deep learning models, this paper proposes a novel explainable artificial intelligence(XAI) method based on the Shapley value. By introducing the Wasserstein ambiguity set into the interpretability framework of KernelSHAP(Kernel SHapley Additive explanation), a distributionally robust Shapley value estimator is developed. Furthermore, we illustrate the conversion of this estimator into a regularized estimator and propose an efficient ADMM(Alternating Direction Method of Multipliers) algorithm to solve the optimization problem. An application study on a real-world industrial process validates the effectiveness of the proposed method.