.png){kind=icon}
Return
Collaborative fuzzy rule-based modeling with a multi-source data environment
.png){kind=icon}
PRE
AI
DOI：10.1016/j.fss.2026.110097.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In real world, a phenomenon is often observed and recorded by multiple institutions. Based on the recorded data, institutions can construct their own predictive models to understand the phenomenon. However, it is often difficult for a single institution to obtain a comprehensive understanding of the complex phenomenon based on its own data. In this study, we select the fuzzy rule-based model as a representative prediction method and introduce how institutions can effectively use multi-source data to construct their own predictive models with full consideration of data privacy. The originality of the study is summarized as that the overlap degree of data sets belonging to different institutions is considered when building collaborative predictive models, and corresponding modeling strategies are designed for scenarios where data sets of different institutions are either with a high (Scenario A) or low (Scenario B) degree of overlap. In Scenario A, the structure of data sets from different sources varies greatly and institutions can take Unionlike strategies to achieve collaboration. For different levels of privacy-retention requirements, we propose strategies based on either sharing data structure or sharing local predictive models. In Scenario B, the structure of data sets from different sources is similar, and individual institutions can take Intersection-like strategies to achieve collaboration. For different levels of efficiency requirements, strategies based on either a single-stage collaboration or a two-stage collaboration are proposed to share local predictive models. Through experiments on a series of synthetic and publicly available data sets, we demonstrate the effectiveness of the proposed approach.
Keywords:
Multi-source data
Collaborative predictive model
Data privacy
Overlap degree of data sets
Journal
IF:
2.7
Papers:
7.6K
Citations:
1.5W
Organization
View more organizations
Cited Papers
No cited papers available