A data reduction method to train, test, and validate neural networks

Abstract

Prediction is an important application of neural networks. When a large data source is used to train a neural network model to make prediction, considerable effort and time are required to obtain reliable outcomes. This paper describes a technique that reduces the size of a large data set but still preserves the pertinent characteristics of the problem domain in the data. Neural network models built using this reduced data set show nearly identical performance on the same set of test cases than models built using the full size data set.