A weighted-sum chaotic sparrow search algorithm for interdisciplinary feature selection and data classification.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Data-driven approaches require efficient processing of large datasets.
• Redundant and non-informative features hinder machine learning (ML) algorithm performance.
• Feature selection (FS) techniques are crucial for optimizing datasets before ML application.

Purpose of the Study:

• To develop an optimized feature selection (FS) technique for machine learning (ML).
• To enhance the Sparrow Search Algorithm (SSA) using chaotic maps for improved performance.
• To address limitations of standard SSA, such as low swarm diversity and weak exploration.

Main Methods:

• A novel wrapper FS technique named Chaotic Sparrow Search Algorithm (CSSA) was developed.
• CSSA integrates ten chaotic maps to improve initial swarm generation, variable substitution, and search range clamping.
• The performance of CSSA was evaluated on benchmark functions and diverse ML datasets.

Main Results:

• CSSA demonstrated superior swarm diversity and convergence speed on IEEE CEC benchmark functions.
• Experimental analysis showed CSSA outperformed twelve state-of-the-art algorithms on UCI and microarray datasets for classification.
• Statistical post-hoc analysis confirmed CSSA's significance in accuracy, feature selection, and stability.

Conclusions:

• CSSA is a highly effective and stable feature selection method.
• The proposed chaotic enhancements significantly improve SSA's exploration and exploitation capabilities.
• CSSA offers a robust solution for optimizing datasets in machine learning applications.