BlindSMOTE: Synthetic Minority Oversampling Based Only on Evolutionary Computation.

Area of Science:

• Data Mining and Machine Learning
• Artificial Intelligence
• Computational Intelligence

Background:

• Class imbalance is a prevalent issue in real-world data mining, where minority classes are underrepresented, negatively impacting classifier performance.
• The Synthetic Minority Oversampling Technique (SMOTE) and its variants are common solutions, but rely on specific hypotheses for synthetic sample generation.
• Existing methods often impose constraints on synthetic instance creation, potentially limiting their effectiveness.

Purpose of the Study:

• To propose a novel class imbalance handling method using evolutionary computation without constraints on synthetic instance generation.
• To integrate majority class undersampling within the evolutionary framework.
• To demonstrate the superiority of the proposed approach over existing strategies.

Main Methods:

• Development of a class imbalance solution based purely on evolutionary computation.
• Generation of synthetic minority class samples without predefined constraints.
• Incorporation of majority class undersampling into the evolutionary process.
• Extensive comparative analysis across multiple datasets and existing strategies.

Main Results:

• The proposed evolutionary computation method demonstrated significant advantages in addressing class imbalance.
• The approach showed improved performance across various classification methods and a large number of datasets.
• The unconstrained synthetic sample generation and integrated undersampling proved effective.

Conclusions:

• Evolutionary computation offers a flexible and powerful framework for tackling class imbalance problems in data mining.
• The proposed method provides a robust alternative to existing SMOTE-based techniques.
• This approach enhances classifier performance on imbalanced datasets through unconstrained synthetic data generation and undersampling.