Stacked ANN ensemble

The current project introduces a script generating a stacked learning ensemble, containing a single multilayered ANN (Level-1 learner or meta-learner) trained using the predictions (meta-data) of a number of ANNs (Level-0 learners).

The Vinyl bromide dataset, that is available with the Deep Learning Toolbox™ was used for training and testing the ensemble. The dataset contains 16 input and 1 output variable and a total of 68308 observations.

Multiple script executions indicated that the stacked ensemble performed 20% better than the best individual ANN.