An efficient hybrid filter-wrapper method based on improved Harris Hawks optimization for feature selection.

Area of Science:

• Machine Learning
• Data Science
• Bioinformatics

Background:

• High-dimensional datasets often suffer from numerous irrelevant features, limited samples, and imbalanced classes.
• These issues significantly degrade the performance of classification algorithms.
• Effective feature selection is crucial for accurate classification in such scenarios.

Purpose of the Study:

• To propose a novel two-stage hybrid method for optimal feature selection.
• To enhance classification performance on high-dimensional, low-sample, and imbalanced datasets.
• To reduce computational costs associated with feature selection.

Main Methods:

• A two-stage approach combining filter and wrapper methods.
• Stage 1: Feature weighting using a filter method to remove irrelevant/redundant features.
• Stage 2: Optimized feature subset identification using an enhanced Harris Hawks Optimization algorithm with GRASP and genetic algorithm operators.

Main Results:

• The proposed algorithm successfully identified an optimal subset of relevant features.
• Experimental results validated the effectiveness of the two-stage hybrid method.

Conclusions:

• The developed two-stage hybrid method efficiently selects optimal feature subsets for high-dimensional data.
• This approach improves classification algorithm performance, particularly with limited and imbalanced samples.
• The method shows promise for diverse applications requiring robust feature selection.