A cluster-based hybrid sampling approach for imbalanced data classification.

Area of Science:

• Machine Learning
• Data Science
• Pattern Recognition

Background:

• Imbalanced datasets pose significant challenges in instrumental data classification.
• Standard classifiers struggle with minority class identification when overwhelmed by majority instances.

Purpose of the Study:

• To introduce a novel hybrid sampling approach, Cluster-based Under-sampling and SMOTE (CUSS), for imbalanced dataset classification.
• To develop a new cluster-based under-sampling technique and a data distribution-based strategy for setting expected instance numbers.

Main Methods:

• CUSS employs a hybrid data-level resampling strategy combining under-sampling and Synthetic Minority Over-sampling Technique (SMOTE).
• A novel cluster-based under-sampling method and a dynamic instance number setting strategy are integral to CUSS.
• The method was evaluated against five other resampling techniques across 15 diverse datasets.

Main Results:

• CUSS demonstrated superior performance in imbalanced dataset classification compared to other state-of-the-art methods.
• Experimental results validated the effectiveness of CUSS across datasets with varying sizes and imbalance ratios.
• The proposed under-sampling and instance number strategies contributed to CUSS's enhanced performance.

Conclusions:

• The Cluster-based Under-sampling and SMOTE (CUSS) approach offers a robust solution for imbalanced instrumental data classification.
• CUSS provides a significant advancement over existing hybrid resampling methods.
• The developed techniques enhance the ability to accurately classify minority classes in imbalanced datasets.