Feature-selected tree-based classification.

Area of Science:

• Machine Learning
• Computer Science
• Pattern Recognition

Background:

• Feature selection enhances classifier performance by reducing noise and redundancy.
• Partially informative features necessitate strategies for decomposing complex classification tasks.
• Binary classifiers often require multiclass problems to be reformulated into two-class subproblems.

Purpose of the Study:

• To develop a novel method for multiclass classification that integrates hierarchical problem decomposition with feature selection.
• To introduce the Feature Selected Hierarchical Classifier (FSHC) algorithm.
• To evaluate the performance of FSHC against established multiclass classification techniques.

Main Methods:

• The proposed method constructs a binary tree of classification subproblems.
• Simultaneous feature selection is performed for each individual classifier within the hierarchy.
• Support Vector Machine (SVM) classifiers are utilized for evaluating the FSHC algorithm.

Main Results:

• The FSHC method demonstrates accuracy comparable to existing common multiclass SVM approaches.
• The FSHC algorithm generates solutions characterized by a reduced number of classifiers.
• The FSHC approach leads to fewer features and decreased testing times compared to other SVM multiclass extensions.

Conclusions:

• The Feature Selected Hierarchical Classifier (FSHC) offers an effective approach for multiclass classification.
• FSHC provides a competitive alternative to existing methods, particularly in terms of efficiency and model simplicity.
• The integration of hierarchical decomposition and feature selection presents a promising direction for advanced classification systems.