Recent research advances on Tangled Program Graphs (TPGs) have demonstrated that Genetic Programming (GP) can be used to build accurate classifiers. However, this performance has been tested on balanced classification problems while most of the real world classification problems are imbalanced, with both over-represented classes and rare classes. This paper explores the effect of imbalanced data on the performance of a TPG classifier, and proposes mitigation methods for imbalance-caused classifier performance degradation using adapted GP selection phases. The GP selection phase is characterized by a fitness function, and by a comparison operator. We show that adapting the TPG to imbalanced data significantly improves the classifier performance. The proposed adaptations on the fitness make the TPG agent capable to fit a model even with 10 4 less examples than the majority class whereas the revised selection phase of the GP process increases the robustness of the method for moderate imbalance ratios.