The study of problem difficulty is an open issue in Genetic Programming (GP). Thegoal of this work is to generate models that predict the expected performance of a GPbasedclassifier when it is applied to an unseen task. Classification problems aredescribed using domain-specific features, some of which are proposed in this work,and these features are given as input to the predictive models. These models arereferred to as predictors of expected performance (PEPs). We extend this approach byusing an ensemble of specialized predictors (SPEP), dividing classification problemsinto specified groups and choosing the corresponding SPEP. The proposed predictorsare trained using 2D synthetic classification problems with balanced datasets. Themodels are then used to predict the performance of the GP classifier on unseen realworlddatasets that are multidimensional and imbalanced. Moreover, as we know, thiswork is the first to provide a performance prediction of the GP classifier on test data,while previous works focused on predicting training performance. Accurate predictivemodels are generated by posing a symbolic regression task and solving it with GP.These results are achieved by using highly descriptive features and including adimensionality reduction stage that simplifies the learning and testing process. Theproposed approach could be extended to other classification algorithms and used asthe basis of an expert system for algorithm selection.< Réduire