Oblique Decision Tree Ensemble via Multisurface Proximal Support Vector Machine.

Area of Science:

• Machine Learning
• Computer Science
• Data Science

Background:

• Traditional decision trees often use orthogonal hyperplanes, limiting their ability to capture complex data relationships.
• Oblique decision trees offer greater flexibility but can be computationally intensive to train.

Purpose of the Study:

• To propose a novel oblique decision tree ensemble method for enhanced classification performance.
• To address the limitations of orthogonal decision boundaries in tree-based classifiers.

Main Methods:

• Utilizes multisurface proximal support vector machines (SVM) to generate oblique hyperplanes in internal nodes.
• Groups training samples into hyper-classes based on geometric properties and random feature subsets.
• Employs regularization techniques to manage small sample sizes during tree growth.

Main Results:

• Demonstrated effectiveness on 44 real-world benchmark classification datasets.
• Achieved superior performance in terms of both computation time and classification accuracy compared to existing methods.
• The proposed oblique decision tree ensemble shows significant advantages in handling complex classification tasks.

Conclusions:

• The novel oblique decision tree ensemble method offers a promising alternative for high-performance classification.
• The integration of multisurface proximal SVM provides a robust mechanism for generating effective oblique decision boundaries.
• This approach shows potential for broad application across various scientific and engineering fields requiring accurate data classification.