Genetic Programming for Automatically Evolving Multiple Features to Classification.

Area of Science:

• Computer Science
• Machine Learning
• Artificial Intelligence

Background:

• High-dimensional data classification faces challenges from large search spaces and complex feature interactions.
• Existing methods like feature selection or construction alone may yield suboptimal feature sets.

Purpose of the Study:

• To investigate the use of genetic programming for simultaneous feature selection and construction.
• To address classification tasks on high-dimensional datasets.

Main Methods:

• Utilized genetic programming for integrated feature selection and construction.
• Tested the approach on 16 diverse datasets.
• Compared performance against seven established feature selection and construction techniques.

Main Results:

• The proposed method significantly increased classification accuracy across datasets.
• Achieved substantial reduction in dataset dimensionality.
• Demonstrated the effectiveness of simultaneously selecting and constructing features.

Conclusions:

• Genetic programming offers a powerful approach for simultaneous feature selection and construction.
• The integrated method overcomes limitations of standalone techniques.
• The generated complementary features lead to improved classification performance and reduced dimensionality.