Imbalanced data classification using graph based transformation.

Area of Science:

• Computer Science
• Machine Learning
• Artificial Intelligence

Background:

• Imbalanced data classification presents significant challenges in real-world applications.
• Existing methods often struggle with skewed class distributions and limited data.

Purpose of the Study:

• To propose an effective feature extraction and classification method for imbalanced image datasets.
• To improve classification performance in scenarios with limited samples and unequal class representation.

Main Methods:

• A two-stage approach involving clustering-based feature extraction (CBFE) for initial feature reduction.
• Graph-based projection using combined Laplacian matrices to preserve class manifold structures.
• Utilizing Support Vector Machine (SVM) as a simple yet effective classifier.

Main Results:

• The proposed method demonstrated superior performance compared to standard SVM and Convolutional Neural Network (CNN) on imbalanced SVHN and CIFAR-10 datasets.
• Achieved better results than CNN even when CNN employed data augmentation, while the proposed method did not use augmentation.
• The method is efficient for both imbalanced data and small sample size classification tasks.

Conclusions:

• The developed method offers a robust solution for image classification with imbalanced data distributions.
• It effectively addresses the limitations of traditional methods, particularly in small sample size scenarios.
• The unsupervised and efficient nature of the proposed technique makes it suitable for practical applications.