.png){kind=icon}
.png){kind=icon}
Return
Graph-driven feature selection via granular-rectangular neighborhood rough sets for interval-valued data sets
.png){kind=icon}
PRE
AI
DOI：10.1016/j.asoc.2025.112716.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In the burgeoning landscape of big data analytics, interval-valued datasets are indispensable for modeling uncertainty and vagueness, with significant implications for sectors such as healthcare and environmental science. Feature selection, a linchpin in data mining, is paramount for streamlining data processing and bolstering predictive models. However, the literature on feature extraction within interval-valued information systems is notably sparse. This paper proposes a groundbreaking feature selection framework that skillfully addresses the complexities of interval-valued data. The method innovatively utilizes a fully connected weighted undirected graph to encapsulate interval data, combining graph-theoretic insights with granular-rectangular neighborhood rough set theory. By evaluating the significance of each attribute based on its importance to the entire information system, and applying matrix power series to accelerate computations, the framework ensures both robust classification performance and the elimination of redundancy, marking a significant advancement in this field. Through comparative experiments on 12 public datasets with 7 other algorithms, theoretical analysis, and experimental results demonstrate that the proposed method not only exhibits high effectiveness in handling interval-valued data but also further improves efficiency and classification performance. In addition, the method also shows significant advantages in reducing the dependence on prior knowledge and improving the interpretability of the model, which fully proves its applicability and reliability in large-scale data analysis.
Keywords:
Feature selection
Graph theory
Interval-valued information system (IVIS)
Matrix power series (MPS)
Neighborhood rough set (NRS)
Journal
IF:
6.6
Papers:
1.4W
Citations:
4.8W