Abstract—In this paper we present a feature selection
algorithm using the idea of importance criteria of a node for a
Deep Neural Network and demonstrate its efficacy in selecting
features for predicting a set of variables in the domain of
computational fluid dynamics. Our algorithm for feature
selection defines an importance criteria for each input feature
based on the weights of a fully trained network from the input
to the first hidden layer and works in two steps: in the first step
it prunes input nodes that are not selected as anchors by any
hidden layer node and in the second step we use an aggregate
weight based importance criteria to further prune the space of
inputs. We use our algorithm for electing sensor locations on a
projectile, such that the sensor data can be used for predicting
the pitch and yaw of the object as it executes different
maneuvers along a complex trajectory. We start with a neural
network which takes all the sensor data as input for predicting
the pitch and yaw. We show that using our algorithm we can
select a subset of the input features without increasing the
training and validation errors significantly.
Index Terms—Neural networks, deep learning, artificial
intelligence, feature selection, computational fluid dynamics.
Chaity Banerjee is with the University of Central Florida, USA (e-mail:
Chaity.BanerjeeMukherjee@ucf.edu).
Tathagata Mukherjee is with the University of Alabama, Huntsville, USA
(e-mail: tathagata.mukherjee@uah.edu).
Chad Lillian is with the Leidos Inc., USA (e-mail:
chad.lillian.ctr@us.sf.mil).
Daniel Reasor and Eduardo Pasiliao Jr. with the Air Force Research
Laboratory, USA (e-mail: daniel.reasor@us.af.mil,
eduardo.pasiliao@us.af.mil).
Xiuwen Liu is with the Florida State University, USA (e-mail:
liux@cs.fsu.edu).
Cite: A Feature Selection Algorithm Using Neural Networks, "A Feature Selection Algorithm Using Neural Networks," International Journal of Machine Learning and Computing vol. 12, no. 1, pp. 31-36, 2022.
Copyright © 2022 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).