Comparative study of variable precision rough set model and graded rough set model

Xianyong Zhang, Zhiwen Mo, Fang Xiong, Wei ChengInt. J. Approx. Reason 2012 · DOI

TL;DR

This paper delves into the nuances of two extended rough set models: the variable precision rough set model (VPRS) and the graded rough set model (GRS). By introducing novel concepts like "rough set regions" and analyzing their properties, the authors provide a comprehensive comparative study, highlighting their strengths, weaknesses, and potential for mutual transformation. This work enriches rough set theory, offering more precise ways to handle uncertainty and imprecision in data analysis.

What Problem Does This Paper Address?

Classical rough set theory, while powerful for dealing with vagueness, has limitations. It strictly defines relationships between equivalence classes and sets, failing to capture the quantitative information about the degree of overlap. This can lead to inaccuracies in applications where such nuances are crucial. To address this, generalized rough set models like VPRS and GRS were developed. VPRS introduces a "precision" parameter to allow for classification errors, while GRS uses a "grade" parameter to account for the degree of overlap. However, a clear understanding of how these two models relate, their comparative advantages, and how they extend classical rough sets remained an area for deeper exploration. This paper aims to bridge that gap by providing a detailed comparative study of VPRS and GRS.

Methods

The paper introduces "rough set regions" as a more precise way to classify the universe within both VPRS and GRS. These regions, including positive, negative, and boundary regions, are derived from the respective approximation operators. The authors then investigate the logical meanings and structural properties of these regions for both models. For the GRS model, they propose and analyze "macroscopic" and "microscopic" algorithms for calculating these regions, comparing their time and space complexities. Furthermore, the paper explores new properties of both models, focusing on how their approximation operators interact with set operations and how they extend classical rough set properties. Finally, it investigates the relationship and potential transformations between the VPRS and GRS models.

Datasets

The paper uses a medical dataset for illustration. This dataset consists of 36 patients with attributes related to fever, headache, and cold. The data is presented in Table 2, and statistical results of patient classes are shown in Table 3. This medical example serves to demonstrate the practical application and interpretation of the rough set regions and algorithms discussed in the paper. No new benchmarks are introduced; rather, an existing dataset is used to exemplify the models.

Key Insights and Ideas

  • Rough Set Regions: The introduction of "rough set regions" (positive, negative, upper/lower boundary) provides a more granular and interpretable way to understand the approximations in both VPRS and GRS, going beyond simple upper and lower approximations.
  • Macroscopic vs. Microscopic Algorithms (GRS): The paper distinguishes between two algorithmic approaches for GRS. The macroscopic algorithm is straightforward but potentially less efficient, while the microscopic algorithm offers better space complexity and can be more efficient in certain scenarios, especially when dealing with large datasets and specific parameter values.
  • Complementary Strengths: VPRS and GRS capture different aspects of uncertainty. VPRS focuses on relative errors (precision), while GRS captures absolute overlap (grade). The paper suggests they can be complementary, especially in situations with significant information gaps between equivalence classes.
  • Extension of Classical Rough Sets: Both VPRS and GRS are shown to be direct extensions of the classical rough set model, with the classical model being a special case when precision or grade parameters are set to specific values (e.g., $\beta=0$ or $k=0$).
  • Idempotency and Limitations: While both models extend classical rough sets, they lose some of the strict idempotency properties of classical approximation operators. The paper identifies specific conditions (e.g., $\beta=1$ in VPRS) that hinder these properties, suggesting the need for "evaluation mappings" to better estimate set relationships.

Key Results

  • Algorithmic Analysis: The paper provides detailed analyses of macroscopic and microscopic algorithms for GRS, showing that while their worst-case asymptotic time and space complexities can be similar ($O(n)$), the macroscopic algorithm has an advantage in time complexity, and the microscopic algorithm has an advantage in space complexity.
  • Medical Example: A medical case study demonstrates how the rough set regions of GRS can provide a composite description of patient classes based on the "grade" quantitative index, showing practical significance. For the example, the time and space complexities of the algorithms were calculated for a small dataset ($n=9$).
  • Properties of VPRS and GRS: The paper systematically derives and presents new properties for both VPRS and GRS, particularly concerning their approximation operators and their interactions with set operations (union, intersection, complement). These properties highlight how the models extend classical rough set theory.
  • Relationship and Transformation: The paper establishes a close relationship between VPRS and GRS, suggesting they can be transformed into each other under certain conditions, particularly by defining "grade k error degree level" and "precision 1-β grade" parameters.

Limitations and Future Work

  • Parameter Sensitivity: The paper notes that the "grade error degree level" and "precision grade" can be integers, rational numbers, or real numbers. However, it also points out that certain conditions (e.g., $|[x]_R| < k$) can lead to parameters exceeding standard ranges, suggesting that graded approximations might be "wider" than variable precision approximations in some senses, but VPRS can be more general due to potentially non-integer parameters.
  • Mutual Transformation Complexity: While a relationship and transformation between the two models are proposed, the paper acknowledges that this transformation has a "local attribute and a specific property." Different equivalence classes might require distinct parameters and confidence levels for a successful transformation, implying that a direct, universal transformation might not be straightforward.
  • Future Work: The authors suggest that further research could explore common binary relations or general approximate spaces, rather than just equivalence partitions. They also emphasize the potential for widespread applications and meaningful combinations of the VPRS and GRS models. The study of rough topology and approximate topology is also mentioned as a future direction.

Summary updated

BibTeX
@article{zhang2012comparative,
  author = {Xianyong Zhang and Zhiwen Mo and Fang Xiong and Wei Cheng},
  title = {{Comparative study of variable precision rough set model and graded rough set model}},
  journal = {Int. J. Approx. Reason.},
  year = {2012},
  doi = {10.1016/j.ijar.2011.10.003},
}

Loading related content...

Figure