Compact Belief Rule Base Learning for Classification with Evidential Clustering.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Belief rule-based classification systems (BRBCS) handle uncertainty but suffer from large belief rule bases (BRB) in big data, hindering interpretability.
• Existing BRBCS models struggle with scalability and maintaining model transparency when dealing with high-dimensional datasets.

Purpose of the Study:

• To propose a novel method for designing a compact belief rule-based classification system (CBRBCS).
• To improve the interpretability of classification models for big data while maintaining comparative accuracy.

Main Methods:

• A supervised evidential C-means clustering (ECM) algorithm is developed for weighted product-space clustering to partition training data.
• A systematic approach constructs belief rules from the obtained credal partitions.
• An evidential partition entropy-based optimization procedure refines the belief rule base (BRB) for compactness.

Main Results:

• The proposed CBRBCS method effectively reduces the size of the belief rule base (BRB).
• Experimental results demonstrate a favorable trade-off between classification accuracy and model interpretability.
• The method achieves comparable accuracy to traditional BRBCS while significantly enhancing transparency.

Conclusions:

• The developed CBRBCS offers a more interpretable classification model for big data applications.
• The integration of evidential C-means clustering provides an efficient way to design compact and accurate belief rule-based systems.
• This approach addresses the critical challenge of interpretability in complex classification tasks involving uncertainty.