A Comparison of Linear Genetic Programming英文电子书.pdf

IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION, VOL. 5, NO. 1, FEBRUARY 2001 17 A Comparison of Linear Genetic Programming and Neural Networks in Medical Data Mining Markus Brameier and Wolfgang Banzhaf Abstract—We introduce a new form of linear genetic program- In contrast to NNs, GP has not been used extensively for med- ming (GP). Two methods of acceleration of our GP approach are ical applications to date. Gray et al. [11] report from an early discussed: 1) an efficient algorithm that eliminates intron code and application of GP in cancer diagnosis, where the results had 2) a demetic approach to virtually parallelize the system on a single processor. Acceleration of runtime is especially important when been found to be better than with an NN. In [16], a grammar- operating with complex data sets, because they are occuring in based GP variant is used for knowledge extraction from medical real-world applications. We compare GP performance on medical databases. Rules for the diagnosis have been derived from the classification problems from a benchmark database with results program tree that uncover relationships among data attributes. obtained by neural networks. Our results show that GP performs The outcomes of different types of classifiers, including NNs comparable in classification and generalization. and genetic programs, are combined in [25]. This strategy re- Index Terms—Data mining, evolutionary computation, genetic sults in an improved prediction of thyroid normal and thyroid programming, neural networks. carcinoma classes.