Using Neural Network for Feature Selection

When building a predictive model, there may be a large number of data fields available for use as inputs to the model. Selecting only those fields most useful to the model has a variety of advantages; it simplifies the model-building process, leading to better and simpler models, and it simplifies the resulting models, leading to more effective insight and easier Deployment.

This Feature Selection can be achieved through a variety of techniques, business and data knowledge can be applied to select the fields likely to be relevant, and univariate techniques can be used to select individual fields that have a relation to the predictive target. It is also a common practice to use other models to help select features whose relevance is more multivariate in nature. Decision trees are often used for this purpose, because building a decision tree model implicitly selects relevant variables; each variable is either used in the model, therefore indicated...