Partial Classifier Chains with Feature Selection by Exploiting Label Correlation in Multi-Label Classification.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Multi-label classification (MLC) assigns multiple labels to an object, requiring effective exploitation of label correlations.
• Existing methods like classifier chains (CC) face challenges with suboptimal label ordering and inclusion of irrelevant features.

Purpose of the Study:

• To propose a novel Partial Classifier Chain with Feature Selection (PCC-FS) method for enhanced multi-label classification.
• To address the limitations of random label ordering and irrelevant feature inclusion in traditional CC methods.

Main Methods:

• PCC-FS integrates feature selection by analyzing label-feature covariance to remove discriminative irrelevant features.
• It simultaneously incorporates coupled labels into the chain structure, optimizing training and prediction.

Main Results:

• Experimental results across five metrics show PCC-FS significantly outperforms eight state-of-the-art MLC algorithms.
• The method effectively exploits label correlations in both label and feature spaces.

Conclusions:

• PCC-FS offers a significant advancement in multi-label classification by simultaneously optimizing label ordering and feature selection.
• The proposed approach enhances predictive performance by effectively managing label dependencies and feature relevance.