Z-Number Generation Model and Its Application in a Rule-Based Classification System

Area of Science:

• Uncertainty Quantification
• Fuzzy Logic and Decision Making
• Data Summarization

Background:

• Z-numbers effectively handle uncertainty and partial reliability in information.
• Existing methods focus on deriving probability distributions from Z-numbers, but not vice-versa.
• The summarization capability of Z-numbers for probability distributions remains an open research question.

Purpose of the Study:

• To propose a novel nonlinear model, Maximum Expected Minimum Entropy (MEME), for generating Z-numbers from sets of probability distributions.
• To introduce Z-valuation if-then rules for classification, enhancing the representation of uncertainty in rule consequents.
• To develop and validate a Z-valuation rule-based (ZVRB) classification system for improved decision-making under uncertainty.

Main Methods:

• Developed the Maximum Expected Minimum Entropy (MEME) nonlinear model to generate Z-numbers directly from data.
• Introduced Z-valuation if-then rules, replacing deterministic consequents with uncertain Z-valuations.
• Implemented a Z-valuation rule-based (ZVRB) classification system.

Main Results:

• The MEME model successfully generates Z-numbers from probability distributions without expert input.
• The ZVRB classification system demonstrated superior classification performance compared to traditional and fuzzy classifiers in experimental evaluations.
• Z-valuation rules effectively summarize uncertain information in classification rule consequents.

Conclusions:

• Z-numbers can effectively summarize sets of probability distributions using the proposed MEME model.
• The ZVRB classification system offers a robust approach for classification tasks involving significant uncertainty.
• The research opens new avenues for applying Z-numbers in data analysis and machine learning.