A three-way efficacy prediction method fusing temporal composite rough set and hybrid machine learning models on multigranulation temporal hybrid attribute information system

Abstract

Efficacy prediction is a key research topic in clinical management practice. To address the general efficacy prediction problem characterized by multigranularity, temporality, and incompleteness, this study proposes a three-way efficacy prediction method that integrates a temporal composite rough set and a hybrid machine learning model (TCRS-HML). First, a multigranulation temporal hybrid attribute information system (MTHAIS) is constructed to handle hybrid attributes exhibiting these characteristics, and the data in MTHAIS is preprocessed using random forest and bag-of-words models. Next, the concept of temporal order is introduced into classical composite rough sets, and temporal equivalence, temporal neighborhood, and temporal similarity relationships are established based on the temporal hybrid attribute matrices of the objects. Subsequently, the definitions of a temporal composite rough set and its attribute reduction method are presented, along with a discussion of their mathematical properties. Finally, efficacy prediction results are obtained by building a hybrid machine learning model pool and selecting the optimal model. Experimental results, based on 493 real temporal medical records from 120 rheumatoid arthritis (RA) patients at the Guangdong Hospital of Traditional Chinese Medicine (2018–2023), show that the accuracy, precision, recall, and F${}_1$ score of the proposed method are 0.8012, 0.8241, 0.8012, and 0.7885, respectively. These results outperform those of 17 comparative methods, demonstrating the scientific validity and feasibility of proposed approach. Furthermore, sensitivity analyses and statistical tests confirm the robustness and generalizability of the method. This study provides a new methodological reference for management science problems such as clinical efficacy prediction and contributes to the integration of rough set theory and machine learning in management and decision sciences.