phiJEN

Feature selection and construction are an important step in pattern recognition, which have gained increased attention in the last years. There are a variety of linear and nonlinear methods, which could be further improved using evolutionary computation. Following on own research the following topics could be addressed in a thesis:

• Evolution of hyperplanes for linear feature construction
• Evolution of artificial neural networks for nonlinear feature construction
• Combination of linear and nonlinear methods
• Construction of class-specific feature sets
• Comparison of wrapper and filter approaches
• Feature constuction in high-dimensional problems