A hybrid machine learning and three-way soft clustering integrated decision-making method with incomplete multi-source heterogeneous attribute information

Abstract

Decision-making under uncertainty in the era of big data and technological innovation can often lead to more objective and scientific results. However, data characterized by multi-source heterogeneity, nonlinearity, imbalance, and incompleteness pose a substantial challenge to traditional decision-making theories and methods. In view of this, this paper defines the concept of an incomplete multi-source heterogeneous attribute information system (IMHAS), introduces an attribute reduction method on IMHAS that integrates rough sets and machine learning, and combines three-way soft clustering and hybrid machine learning models. A novel theoretical framework is proposed to address uncertain decision-making problems involving data characterized by multi-source heterogeneity, nonlinearity, imbalance, and incompleteness is proposed. First, IMHAS is established and its attribute reduction method is defined using rough set, neighborhood rough set, bag-of-words, and random forest techniques. Second, to further resolve the correlation between objects in IMHAS, a three-way soft clustering method is introduced. Finally, to target decision-making for different object categories, two types of machine learning models are constructed for handling discrete and conditional decision attributes. The scientific superiority of the proposed method was verified using four distinct datasets from medical and healthcare domains. The results show that the proposed method outperforms all comparative methods, and it can effectively support uncertain decision-making in four different cases. In conclusion, this paper proposes a general theoretical framework for uncertain decision-making based on artificial intelligence techniques for incomplete multi-source heterogeneous data, thus offering realistic guidance for clinical applications involving such data.