LI-MLC: a label inference methodology for addressing high dimensionality in the label space for multilabel...

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Multilabel classification (MLC) is increasingly researched for real-world applications.
• Existing methods transform data or adapt algorithms but struggle with many labels.
• High dimensionality in the label space is a significant challenge in MLC.

Purpose of the Study:

• To analyze label space dimensionality in multilabel data sets (MLDs).
• To introduce a transformation methodology using association rules for MLC.
• To reduce the label space and improve MLC algorithm efficiency.

Main Methods:

• Utilizing association rules to discover dependencies between labels in MLDs.
• Transforming MLDs by reducing the label space based on discovered dependencies.
• Employing a postprocessing stage to infer deleted labels.

Main Results:

• The proposed methodology was validated through extensive experimentation.
• Performance improvements were statistically significant in certain cases.
• The approach effectively handles MLDs with a large number of labels.

Conclusions:

• Label space dimensionality is a critical issue in MLC.
• Association rule-based transformation offers a viable solution for high-dimensional MLDs.
• The method enhances MLC algorithm performance and efficiency.