A Generic multi-dimensional feature extraction method using multiobjective genetic programming.

Area of Science:

• Machine Learning
• Pattern Recognition
• Computational Intelligence

Background:

• Traditional pattern classification often requires extensive feature engineering and classifier selection.
• Optimizing feature extraction and selection simultaneously remains a challenge in machine learning.

Purpose of the Study:

• To present a generic feature extraction method for pattern classification using multiobjective genetic programming.
• To evolve vector-to-vector feature extractors that maximize class separability while optimizing decision space dimensionality.

Main Methods:

• Utilized multiobjective genetic programming to evolve feature extractors.
• Developed a framework that simultaneously optimizes feature mappings and decision space dimensionality.
• Performed statistically-founded comparisons against established classifier paradigms.

Main Results:

• The proposed evolutionary method achieved statistically smaller misclassification errors compared to most established classifiers.
• In worst-case scenarios, the method showed no statistical difference in performance against comparators.
• Both feature extraction and feature selection were identified as crucial components for the technique's success.

Conclusions:

• The presented method effectively performs feature extraction and selection, leading to improved classification accuracy.
• This approach obviates the need for exhaustive evaluation of conventional classifiers for new pattern recognition problems.
• The evolutionary technique offers a robust and efficient alternative for enhancing pattern classification performance.