A Hybrid Feature Selection Method Based on Binary State Transition Algorithm and ReliefF.

Area of Science:

• Data Mining
• Machine Learning
• Bioinformatics

Background:

• Feature selection is crucial in data mining but challenging due to NP-hard complexity.
• Existing methods may struggle with local optima and efficient feature subset evaluation.

Purpose of the Study:

• To propose a novel, efficient hybrid feature selection method, ReliefF-BSTA.
• To improve classification accuracy in data mining and biomedical applications.

Main Methods:

• ReliefF-BSTA combines filter and wrapper phases for feature selection.
• It employs an initialization approach based on feature ranking and a probability substitute operator.
• A k-nearest neighborhood classifier with leave-one-out cross-validation is used for evaluation.

Main Results:

• ReliefF-BSTA demonstrated superior efficiency in classification accuracy compared to other methods.
• Average accuracy improved by approximately 2.5% on public datasets versus filter methods.
• Significant accuracy increase from 77.24% to 85.25% on the AID1284 biomedical dataset.

Conclusions:

• The proposed ReliefF-BSTA method is effective for feature selection in data mining.
• It offers advantages in handling local optima and updating solutions based on feature weights.
• ReliefF-BSTA shows significant potential for improving classification performance on diverse datasets.