A decision-making approach under uncertainty based on ensemble learning model with multimodal data and its application in medical diagnosis
Xixuan Zhao,Bingzhen Sun,Xiaoli Chu,Darong Wu,Guli Jiang,Hui Zhou,Jianxiong Cai
DOI: https://doi.org/10.1016/j.eswa.2024.125983
IF: 8.5
2024-12-11
Expert Systems with Applications
Abstract:Incorporating data with multimodal attributes such as numerical, category and text together into decision-making information can effectively improve decision-making accuracy. The ways of construct a decision-making information system through multimodal data and its application in practice is a cutting-edge scientific problem. Based on the multimodal data decision-making information system (MDIS), neighborhood rough set (NRS), clique percolation method (CPM) and ensemble learning model, this paper proposed a general theoretical framework for solving the decision-making problems under uncertainty with the characteristics of data multimodality, nonlinearity and sample imbalance. First, MDIS was constructed based on the multimodal data consisting of numerical, category and textual attributes. Furthermore, the NRS model on MDIS was constructed by defining the distance metric method and the neighborhood object set construction method. In order to effectively obtain the valuable information in MDIS, the basic theory of NRS was systematically studied and attribute reduction was performed at the same time. For further analyzing the correlation between the objects in MDIS, CPM was used to classify the attribute-reduced objects into communities. Namely, the objects were clustered. Finally, seven machine learning classifiers and soft voting ensemble learning models were constructed based on support vector machines, K -nearest neighbors, decision trees and naive bayes, so as to make decisions about the objects in different communities. In this paper, the scientific validity and superiority of the proposed methodology was verified. Specifically, the data of 5632 dampness syndrome patients, 151 colorectal cancer patients, and 32 rheumatoid arthritis patients in The Second Affiliated Hospital of Guangzhou University of Chinese Medicine and Guangzhou No. 11 People's Hospital were used as examples to conduct the empirical study and give consistent explanations with the real clinical practice. The main contribution of this paper is twofold. First, it proposes a theoretical framework for general decision-making problems under uncertainty characterized by data multimodality, nonlinearity, and sample imbalance. Second, it provides a reference for the general clinical decision-making problems to formulate a combined integrative medicine intervention decision.
computer science, artificial intelligence,engineering, electrical & electronic,operations research & management science