Multi-label Random Subspace Ensemble Classification.

Area of Science:

• Machine Learning
• Data Science
• Computational Statistics

Background:

• Multi-label classification presents challenges in assigning multiple labels to data instances.
• Existing ensemble methods may not optimally handle high-dimensional feature spaces inherent in multi-label problems.

Purpose of the Study:

• To develop a novel ensemble learning framework, multi-label Random Subspace Ensemble (mRaSE), for improved multi-label classification.
• To introduce iterative and model-free extensions (Super mRaSE) for enhanced performance and flexibility.
• To provide a model-free feature ranking mechanism compatible with various base classifiers.

Main Methods:

• mRaSE employs random subspace sampling, selecting optimal subspaces based on cross-validation error.
• The framework aggregates selected weak learners to form a robust multi-label classifier.
• Iterative refinement and a Super mRaSE extension incorporating multiple base classifiers are developed.

Main Results:

• The proposed mRaSE algorithms demonstrate superior prediction performance compared to state-of-the-art methods like random forest and deep neural networks.
• Extensive simulations and real-world data applications validate the effectiveness of mRaSE and Super mRaSE.
• The algorithms provide reliable model-free feature ranking.

Conclusions:

• mRaSE offers a powerful and flexible approach to multi-label classification with enhanced predictive accuracy.
• The developed extensions, including Super mRaSE, further advance the capabilities for complex multi-label tasks.
• The R package RaSEn provides an accessible implementation of these advanced ensemble learning algorithms.