A comparison of classification strategies in genetic programming with unbalanced data

Abstract

Machine learning algorithms like Genetic Programming (GP) can evolve biased classifiers when data sets are unbalanced. In this paper we compare the effectiveness of two GP classification strategies. The first uses the standard (zero) class-threshold, while the second uses the "best" class-threshold determined dynamically on a solution-by-solution basis during evolution. These two strategies are evaluated using five different GP fitness across across a range of binary class imbalance problems, and the GP approaches are compared to other popular learning algorithms, namely, Naive Bayes and Support Vector Machines. Our results suggest that there is no overall difference between the two strategies, and that both strategies can evolve good solutions in binary classification when used in combination with an effective fitness function. © 2010 Springer-Verlag.