Classification as clustering: a Pareto cooperative-competitive GP approach

Area of Science:

• Artificial Intelligence
• Machine Learning
• Evolutionary Computation

Background:

• Population-based algorithms like genetic programming naturally support task decomposition among multiple individuals.
• Existing methods often require pre-specifying the number of cooperating individuals in a team.

Purpose of the Study:

• To present a framework for evolving teams of cooperating individuals without pre-specifying team size.
• To enable individuals to learn unique subsets of a classification task.

Main Methods:

• Evolving individuals that map to outcome distributions, which parameterize local membership functions after clustering.
• Utilizing evolutionary multiobjective optimization (EMO) for accurate and non-overlapping behaviors.
• Employing Pareto competitive coevolution for scalability with large, unbalanced datasets.

Main Results:

• The proposed framework balances classification performance and model complexity.
• Effectiveness is demonstrated across 12 UCI datasets with varying instance counts.
• Performance is competitive with nonlinear Support Vector Machine (SVM) classifiers, especially on large datasets.

Conclusions:

• The coevolutionary multiobjective genetic programming framework offers an effective approach to team evolution for classification.
• The method successfully handles task decomposition and scales to large datasets.
• This approach provides a flexible alternative to fixed-size team structures in evolutionary computation.