Hybrid extreme rotation forest.

Area of Science:

• Machine Learning
• Ensemble Learning
• Classification Algorithms

Background:

• Ensemble learning methods enhance predictive accuracy by combining multiple models.
• Extreme Learning Machines (ELM) offer efficient training for neural networks.
• Integrating diverse learning paradigms can lead to improved classification performance.

Purpose of the Study:

• To introduce the Hybrid Extreme Rotation Forest (HERF), a novel ensemble learning algorithm.
• To evaluate the effectiveness of HERF against state-of-the-art ensemble methods.
• To investigate the impact of different data rotation techniques and training strategies.

Main Methods:

• The HERF algorithm combines Decision Trees with Extreme Learning Machines (ELM).
• Each classifier in HERF is trained on randomized data rotations (PCA or Quartimax).
• The study compares HERF with other ensemble methods like Voting ELM and Random Forest.

Main Results:

• HERF demonstrated significant improvements in classification accuracy across benchmark datasets.
• Data rotation using Quartimax outperformed Principal Component Analysis (PCA).
• The HERF approach showed relative insensitivity to regularization parameters.

Conclusions:

• HERF represents a significant advancement in ensemble learning for classification tasks.
• The proposed data rotation strategy and integration of ELM with Decision Trees are effective.
• HERF offers a robust and high-performing alternative to existing ensemble methods.