Boosting Improves Stability and Accuracy of Genetic Programming in Biological Sequence Classification

Abstract

Biological sequence analysis presents interestingchallenges for machine learning. Using one of the mostimportant current problems -- the recognition offunctional target sites for microRNA molecules -- as anexample, we show how joining multiple geneticprogramming classifiers improves accuracy and stabilitytremendously. When moving from single classifiers tobagging and boosting with cross validation andparameter optimisation, you require more computingpower. We use a special-purpose search processor forfitness evaluation, which renders boosted geneticprogramming practical for our purposes.