EPX: An R package for the ensemble of subsets of variables for highly unbalanced binary classification

Area of Science:

• Machine Learning
• Bioinformatics
• Computational Chemistry

Background:

• Binary classification with rare classes presents challenges due to limited data for the minority class.
• High-dimensional feature spaces, common in drug discovery and bioinformatics, exacerbate this issue.
• Standard classification models struggle to leverage rich feature information with sparse rare-class data.

Purpose of the Study:

• To develop an R package for creating an ensemble classifier to detect rare class objects.
• To address the challenge of limited information for rare classes in high-dimensional datasets.
• To improve the exploitation of feature variable richness in classification tasks.

Main Methods:

• The Ensemble of Phalanxes (EPX) algorithm clusters feature variables into diverse subsets (phalanxes).
• These phalanxes are aggregated to form a powerful ensemble classifier.
• Parallel computing is utilized to manage computational load with high-dimensional data.

Main Results:

• The EPX R package implements the Ensemble of Phalanxes algorithm.
• EPX demonstrates superior performance in predictive ranking of rare classes compared to existing methods.
• Parallel computing integration facilitates efficient processing of high-dimensional datasets.

Conclusions:

• The EPX R package offers a flexible approach to clustering feature variables for ensemble classification.
• It effectively ranks rare class objects in highly unbalanced binary classification problems.
• EPX is valuable for applications like drug discovery and protein homology detection, and is available on CRAN.