21. Finding Relevant Variables and Interactions in Neural Network Credit Scoring Models

—Rudy Setiono, School of Computing, National University of Singapore


Credit scoring data contain both discrete and continuous variables that are used in building models to differentiate between good and bad credit risks. Neural networks (NN) are known to produce good results when applied to analyze these types of data. In this paper, we propose the introduction of interaction terms among the continuous variables to improve the NN accuracy and to simplify the classification rules extracted from the neural networks. In addition to the interaction terms, discretized values of the continuous variables are also added before training the neural networks. ...

Get Trends and Research in the Decision Sciences: Best Papers from the 2014 Annual Conference now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.