A new Monte Carlo sampling method based on Gaussian Mixture Model for imbalanced data classification.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Imbalanced data classification is a significant challenge in machine learning.
• Existing methods often fail to account for the complexity of imbalanced distributions, leading to biased models.
• Current data-level and algorithm-level techniques may overemphasize certain difficult factors in minority data.

Purpose of the Study:

• To propose a novel Monte Carlo sampling algorithm based on Gaussian Mixture Model (MCS-GMM) for imbalanced data classification.
• To generate synthetic data that better represents the underlying distribution of imbalanced datasets.
• To mitigate the impact of data overlap and improve the learning process for minority classes.

Main Methods:

• Fitting the imbalanced data distribution using a Gaussian Mixture Model (GMM).
• Applying the Monte Carlo algorithm to generate new synthetic data points.
• Utilizing the three sigma rule to categorize data and reduce overlap.
• Assigning instance weights based on neighbor proximity and probability density functions.

Main Results:

• The proposed MCS-GMM method demonstrates effectiveness in handling imbalanced datasets.
• Experimental results show superior performance compared to existing methods like Synthetic Minority Over-sampling TEchnique (SMOTE).
• The approach successfully generates samples in complex regions, addressing limitations of prior techniques.

Conclusions:

• MCS-GMM offers a robust solution for imbalanced data classification.
• The method enhances model generalization by creating more informative synthetic data.
• This technique provides a valuable advancement in addressing data imbalance challenges in machine learning.