Integrating new data balancing technique with committee networks for imbalanced data: GRSOM approach

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Imbalanced datasets pose significant challenges in classification tasks.
• Existing data balancing techniques like SMOTE and RT have limitations.
• Committee networks offer potential for improved classification performance.

Purpose of the Study:

• To propose a novel data balancing technique for imbalanced classification problems.
• To integrate the proposed technique with a committee network architecture.
• To evaluate the effectiveness of the proposed method against established techniques.

Main Methods:

• A data balancing technique based on the Growing Ring Self-Organizing Map (GRSOM) unsupervised learning algorithm.
• GRSOM generates synthetic data while preserving original data topology.
• Performance evaluation using four UCI Machine Learning Repository datasets and fivefold cross-validation.

Main Results:

• The proposed GRSOM-based data balancing technique, when used with a committee network, performs at least as well as Minority Over-Sampling Technique (SMOTE) and Random Under-sampling Technique (RT).
• Classifiers utilizing neural networks with the backpropagation algorithm demonstrated greater robustness compared to support vector machines.
• The GRSOM method effectively balances data by growing new samples on a ring structure, maintaining data topology.

Conclusions:

• The GRSOM-based data balancing technique is a viable and effective method for imbalanced classification.
• Committee networks combined with GRSOM offer a competitive alternative to existing balancing methods.
• Neural networks with backpropagation show superior robustness in imbalanced classification scenarios compared to SVMs.