Evolutionary Linear Discriminant Analysis for Multiclass Classification Problems

This paper implements Linear Discriminant Analysis (LDA) together with genetic programming symbolic classification for financial multiclass classification problems. A very brief description of the matrix theory of LDA is included. The implementation details in an industrial strength multiclass classification system are discussed. The algorithm is tested using statistically correct, out of sample training and testing. The algorithm's behavior is demonstrated on real world problems previously published as UCI test suites and financial real world problems.