Using machine learning on feature selection
A technical analysis on combining features and their knockoffs within a recurrent neural network
Feature selection, or variable selection, is an important task that is needed within a lot of disciplines. This study applies deep learning on the selection of features using a proposed model. This study contributes to existing literature by examining the performance of a recurrent neural network in a model where both knockoffs of features, or almost identical copies of variables, and the true features are combined within one model framework. This model is used for variable selection purposes where a controlled false discovery rate is used. A simulation study shows that adding additional information with a recurrent neural network does not seem to improve a selection of a subset of features and that combining both knockoff features and true features does only give an advantage to subset selection in case of a feedforward neural network.
|Keywords||Neural network, feedforward neural network, recurrent neural network, Elman net-, work, machine learning, deep learning, DeepPINK, model-X knocknoffs, false discovery rate, feature selection, simulation.|
|Thesis Advisor||Pick, A.|
Biharie, R.A. (2020, May 11). Using machine learning on feature selection. Econometrie. Retrieved from http://hdl.handle.net/2105/52042