In stacked generalization, several algorithms are trained on the training set (i.e. at layer 1) and their predictions are then stacked using a layer 2 model. In many documentations, it is said that it is better that the layer 1 algorithms should be of low correlation. How can one compute this correlation between algorithms ?
For regression tasks correlation will be simply the correlation between the predicted values, for binary classification it will be correlation between predicted probabilities. In multiclass classification you can find correlation between predicted factor variables using the
hetcor package in R
I'm not aware of a simple way to compare. I've more read that you want a diverse set across different types of algorithms to reduce "group think", so you would choose an SVM, a NN, a decision tree, etc. If at layer 2 they tend to vote together, that either means they were all fooled for the same reason, or you found predictable data.