GRADIENT: Grammar-driven genetic programming framework for building multi-component, hierarchical predictive systems

This work presents the GRADIENT (GRAmmar-DrIven ENsemble sysTem) framework for the generation of hybrid multi-level predictors for function approximation and regression analysis tasks. The proposed model uses a context-free grammar guided genetic programming for the automatic building of multi-component prediction systems with hierarchical structures. A multi-population evolutionary algorithm together with resampling and cross-validatory approaches are used to increase component models’ diversity and facilitate more robust and efficient search for accurate solutions. The system has been tested on a number of synthetic and publicly available real-world regression and time series problems for a range of configurations in order to identify and subsequently illustrate and discuss its characteristics and performance. GRADIENT has been shown to be very competitive and versatile when compared to a number of state-of-the-art prediction methods.

► We present a grammar based evolutionary framework for the generation of ensembles.
► We generate hybrid predictors for function approximation and regression.
► Ensembles have hierarchical structures, using a variety of base predictors.
► We use a multi-population evolutionary algorithm with resampling capabilities.