Paffenroth, Randy Clinton
This project combines feedforward neural networks (FFNN) with ensemble learning in a classification model applied to credit card default. To solve this problem, we generate FFNNs as the members of the ensemble. Each FFNN in the ensemble is built using different subsets of predictors as inputs and a separate predictor as the target (instead of the label). The output from the final hidden layer of each network is aggregated into a new dataset. A classification model then uses this new dataset to predict credit defaults. The classification model is evaluated using various metrics to show that an increase in ensemble size increases the consistency of the performance of the model.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.