Study of Complex-valued Learning algorithms for Post-surgery survival prediction

Abstract

Prediction of post-surgery survival of breast cancer patients is critical for long term medical care. In this paper, we study the performances of several complex-valued classifiers in predicting the post-surgical survival, based on the real world Haber data set available in the UCI machine learning repository. The complex-valued classifiers used in the study include the Fully Complex-valued Radial Basis Function (FC-RBF), Fully Complex-valued Relaxation Network (FCRN), Meta-cognitive Fully Complex-valued Relaxation Network (McFCRN), Fully Complex-valued Fast Learning Classifier (FC-FLC), Meta-cognitive Fully Complex-valued Fast Learning Classifier (Mc-FCFLC), Fully Complex-valued Functional Link Network (FCFLN), and Meta-cognitive Fully Complex-valued Functional Link Network (Mc-FCFLN). As the classification performance of the complex-valued classifiers is boosted by the presence of orthogonal decision boundaries, all these classifiers perform better than the state-of-the-art real-valued classifiers. Performance results also show that the Mc-FCFLC and McFCRN outperform other classifiers used in the study. This can be attributed to the meta-cognition that helps in strategic learning in these classifiers.