Is it ok to use predictions from a model on training set, to use it as feature, and train the model again.
I have mixed thoughts on this.
I think the model will get highly dependent on the previous predictions.
Any thoughts are welcome.
Hi shadab,
No using the previous prediction as a feature for the next prediction is not a new idea. This is the bases of LSTM models where the previous word prediction is used for predicting the next word.
But is it feaseable for boosting models,
I have used the approach , to use predictions on training set to add as a feature, model accuracy has increased but , the model has become highly dependent on that particular predictions.