A novel hierarchical selective ensemble classifier with bioinformatics application.

Area of Science:

• Machine Learning
• Bioinformatics
• Computational Biology

Background:

• Selective ensemble learning improves model generalization by selecting diverse, accurate base models.
• High-dimensional data presents challenges for traditional machine learning algorithms.
• Accurate classification is crucial for various bioinformatics applications.

Purpose of the Study:

• To propose a novel selective ensemble learning algorithm, Parallel Optimization and Hierarchical Selection (PTHS).
• To introduce a new feature selection method, Maximize the Sum of Relevance and Distance (MSRD), for high-dimensional data.
• To demonstrate the effectiveness and robustness of PTHS in bioinformatics.

Main Methods:

• PTHS algorithm utilizes parallel optimization and hierarchical selection with k-means for candidate model pruning.
• MSRD feature selection method maximizes relevance and distance to handle high dimensionality.
• Majority voting with a divide-and-conquer strategy is used to combine base model predictions, optimizing computation time.
• The algorithm transforms multi-class problems into binary classification for broader applicability.

Main Results:

• MSRD effectively addresses the challenges posed by high-dimensional data.
• PTHS demonstrates superior performance compared to existing classification algorithms.
• The PTHS classifier achieved high performance in bioinformatics tasks like tRNA identification and protein-protein interaction prediction.

Conclusions:

• PTHS is an efficient and robust selective ensemble learning algorithm.
• The MSRD feature selection method is effective for high-dimensional datasets.
• PTHS shows significant potential for advancing classification tasks in bioinformatics.