Genetic programming for stacked generalization

In machine learning, ensemble techniques are widely used to improve the performance of both classification and regression systems. They combine models generated by different learning algorithms, typically trained on data subsets or with parameters, obtain more accurate models. Ensemble strategies range from simple voting rules complex effective stacked approaches. based adopting a meta-learner, i.e. further algorithm, predictions provided single algorithms making up ensemble. The paper aims at exploiting some most recent genetic programming advances in context generalization. particular, we investigate how evolutionary demes despeciation initialization technique, ϵ-lexicase selection, geometric-semantic operators, semantic stopping criterion, can be effectively GP-based systems’ for generalization (a.k.a. stacking). experiments, performed broad set synthetic real-world problems, confirm effectiveness proposed approach.