Prototype Generation Using Multiobjective Particle Swarm Optimization for Nearest Neighbor Classification.

Area of Science:

• Machine Learning
• Computer Science

Background:

• Nearest Neighbor (NN) classification faces high time complexity due to exhaustive training set searches.
• Prototype generation offers a solution by reducing the data needed for classification.

Purpose of the Study:

• To apply particle swarm optimization (PSO) for effective prototype generation.
• To introduce novel methods for enhancing NN classifier performance and mitigating overfitting.

Main Methods:

• Utilized particle swarm optimization (PSO) for prototype generation.
• Introduced an 'error rank' fitness function considering misclassified instance ranks.
• Implemented a multiobjective (MO) optimization strategy for simultaneous performance on data subsets.

Main Results:

• The proposed PSO-based prototype generation significantly improved NN classification performance.
• The 'error rank' and MO optimization strategies effectively enhanced classification ability and reduced overfitting.
• Experimental results demonstrated superior performance compared to approximately 30 existing prototype generation algorithms across 90 datasets.

Conclusions:

• PSO-based prototype generation with novel fitness functions and MO strategies offers a powerful approach to improve NN classifiers.
• The proposed methods provide a robust solution for reducing classification time complexity and enhancing accuracy.
• This work advances prototype generation techniques in machine learning.