New Genetic Programming Hypothesis Search Strategies for Improving the Interpretability in Medical Data Mining Applications

In this paper we describe a new variant of offspring selection applied to medical diagnosis modeling which is designed to guide the hypothesis search of genetic programming towards more compact and more easy to interpret prediction models. This new modeling approach aims to combat the bloat phenomenon of genetic programming and is evaluated on the basis of medical benchmark datasets. The classification accuracies of the achieved results are compared to those of published results known from the literature. Regarding compactness the models are compared to genetic programming prediction models achieved without the new offspring selection variant.