Social coevolution and Sine chaotic opposition learning Chimp Optimization Algorithm for feature selection

Area of Science:

• Machine Learning
• Swarm Intelligence
• Optimization Algorithms

Background:

• Feature selection is crucial in machine learning, with swarm intelligence algorithms offering potent optimization capabilities.
• The Chimp Optimization Algorithm (CHoA) is recognized for its speed and simplicity but struggles with balancing exploration and exploitation, leading to premature convergence.
• These limitations hinder CHoA's effectiveness in complex feature selection scenarios.

Purpose of the Study:

• To address the limitations of the standard Chimp Optimization Algorithm (CHoA) in feature selection.
• To introduce an enhanced algorithm, the Social coevolution and Sine chaotic opposition learning Chimp Optimization Algorithm (SOSCHoA), to improve optimization accuracy and prevent premature convergence.
• To evaluate the performance of SOSCHoA on high-dimensional classification datasets.

Main Methods:

• The proposed Social coevolution and Sine chaotic opposition learning Chimp Optimization Algorithm (SOSCHoA) integrates social coevolution for improved local search.
• Sine chaotic opposition learning is incorporated to enhance population diversity and mitigate local optima entrapment.
• The algorithm's efficacy was tested through extensive experiments on 12 high-dimensional classification datasets.

Main Results:

• SOSCHoA demonstrated superior performance compared to existing algorithms across multiple metrics.
• The enhanced algorithm achieved higher classification accuracy on the tested datasets.
• SOSCHoA exhibited improved convergence speed and enhanced stability in feature selection tasks.

Conclusions:

• The Social coevolution and Sine chaotic opposition learning Chimp Optimization Algorithm (SOSCHoA) effectively addresses the limitations of the standard CHoA for feature selection.
• SOSCHoA offers significant advantages in classification accuracy, convergence, and stability, particularly for high-dimensional datasets.
• Future research can focus on further optimizing SOSCHoA for feature dimensionality reduction.