A Random Forests Quantile Classifier for Class Imbalanced Data.

Area of Science:

• Machine Learning
• Statistics
• Data Science

Background:

• Class imbalance is a common challenge in machine learning.
• Existing methods for class imbalance may not simultaneously optimize key performance metrics.

Purpose of the Study:

• To propose the q*-classifier, a novel approach for addressing class imbalance.
• To demonstrate the dual optimization properties of the q*-classifier.

Main Methods:

• The q*-classifier assigns samples to the minority class based on conditional probability thresholds.
• The method is motivated by a density-based approach and linked to cost-weighted Bayes classifiers.
• Random forests are utilized for implementing the q*-classification, termed RFQ.

Main Results:

• The q*-classifier maximizes the sum of true positive and true negative rates.
• It minimizes weighted risk, achieving near-zero risk in imbalanced problems.
• RFQ demonstrates competitive or superior performance compared to existing techniques in terms of mean performance and variable selection.

Conclusions:

• The q*-classifier offers a robust solution for class imbalance problems.
• RFQ provides an effective implementation of the q*-classifier using random forests.
• The method shows promise for multiclass imbalanced settings.