Improving Gaussian Naive Bayes classification on imbalanced data through coordinate-based minority feature mining.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Gaussian Naive Bayes (GNB) classifiers struggle with imbalanced datasets, leading to performance degradation.
• Existing sampling techniques for imbalanced data alter data distribution and can cause overfitting or class overlap.
• There is a need for methods that improve GNB performance on imbalanced data without modifying the original dataset.

Purpose of the Study:

• To propose a novel coordinate transformation algorithm based on radial local relative density changes (RLDC).
• To enhance GNB classification performance on imbalanced datasets by generating new features.
• To preserve the original data's quantity and distribution while improving minority class representation.

Main Methods:

• Developed a coordinate transformation algorithm that converts absolute coordinates to RLDC-relative coordinates.
• The RLDC transformation reveals latent local relative density change features, highlighting minority class patterns.
• Applied the transformed features to the GNB classifier to improve minority class probability estimation.

Main Results:

• The RLDC-based coordinate transformation algorithm significantly improved GNB performance on 20 imbalanced datasets.
• The algorithm outperformed 14 traditional sampling algorithms across three classification evaluation metrics.
• Achieved average performance improvements of 21.84%, 33.45%, and 54.63% compared to existing methods.

Conclusions:

• The RLDC coordinate transformation offers a novel and effective approach to handling imbalanced data in GNB classification.
• This method enhances classification accuracy by creating informative features without altering the original data.
• The algorithm demonstrates significant theoretical and practical value for imbalanced classification problems.