A new incomplete pattern classification method based on evidential reasoning.

Area of Science:

• Computer Science
• Machine Learning
• Artificial Intelligence

Background:

• Classifying incomplete patterns is challenging due to missing data, leading to classification uncertainty.
• Existing methods struggle to manage ambiguity arising from multiple potential data imputations.

Purpose of the Study:

• To propose a new prototype-based credal classification (PCC) method for handling incomplete patterns.
• To effectively manage classification uncertainty caused by missing data using evidential reasoning.

Main Methods:

• Utilizes a belief function framework and class prototypes to estimate missing values.
• Employs a credal combination method to integrate multiple classification results from different estimations.
• Assigns difficult-to-classify patterns to meta-classes to reduce errors.

Main Results:

• The PCC method successfully classifies incomplete patterns by combining distinct classification outcomes.
• The credal combination method effectively characterizes uncertainty from conflicting estimations.
• Demonstrated effectiveness through experiments on artificial and real datasets.

Conclusions:

• The proposed PCC method offers a robust approach to classifying incomplete patterns with inherent uncertainty.
• This method improves classification accuracy by leveraging evidential reasoning and a novel combination technique.