面向数据驱动的多准则分类的偏好学习方法及应用研究

随着新一代信息技术的飞速发展，数据驱动的决策正在重新定义管理决策的过程和方式，而传统的多准则分类(MCS)方法已经难以处理数据驱动的决策问题。本项目根据决策要素和流程的变化，分别从特征选择、偏好模型构建、复杂准则处理、算法实现和方法应用等方面研究了数据驱动的MCS问题，并在方法层面提出具有针对性的偏好学习方法，包括基于特征选择的偏好学习方法、基于正则化框架的偏好学习方法、考虑交互准则的偏好学习方法、考虑非单调准则的偏好学习方法。同时，本项目将所提数据驱动的偏好学习方法应用于实际问题，并验证方法的合理性和稳健性。本项目的研究成果将为开展数据驱动的决策提供理论依据和实践指导。

With the rapid advancement of new information technology, data-driven decision-making has redefined the process and ways for managerial decision. However, traditional multiple criteria sorting (MCS) cannot deal with such problems. According to the new changes in the elements and process in decision-making, this project will study data-driven MCS problems from the aspects of feature selection for decision-making, preference model construction, complex criteria processing, algorithm implementation and real-world applications, and propose corresponding preference learning approaches, including the ones based on feature selection, the ones based on the regularization framework, the ones for dealing with interacting criteria, and the ones for handling non-monotonic criteria. Furthermore, the proposed preference learning approaches will be applied to address real-world problems, and their validity and robustness will be studied. The research outcomes of this project will provide theoretical foundations and practical guidance for data-driven decision-making practice.