Naive Bayes-guided bat algorithm for feature selection

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• The exponential growth of data necessitates efficient feature selection methods.
• Improved feature selection positively impacts pattern recognition, machine learning, and signal processing.

Purpose of the Study:

• To introduce a bio-inspired feature selection algorithm, the Bat Algorithm hybridized with a Naive Bayes classifier (BANB).
• To evaluate BANB's performance against existing algorithms on diverse benchmark datasets.

Main Methods:

• Hybridization of the Bat Algorithm with the Naive Bayes classifier.
• Performance evaluation across twelve benchmark datasets, assessing feature count, accuracy, stability, and generalization.
• Comparison with three established feature selection algorithms.

Main Results:

• BANB significantly reduced the number of selected features compared to other methods.
• The algorithm maintained high classification accuracy while removing irrelevant, redundant, or noisy features.
• BANB demonstrated superior stability and produced more generalizable feature subsets.

Conclusions:

• The proposed BANB algorithm is an effective and stable feature selection method.
• BANB offers advantages in data preprocessing for machine learning tasks.
• The approach successfully addresses the challenges posed by large datasets.