Abstract

In our previous study, we introduced two major applications for developing advanced deep learning methods that were focused on credit card data analysis. Herein, we validated our proposed methods by comparing benchmark experiments
with other machine learning approaches. Through these experiments, we confirmed that deep learning exhibits accuracy similar to that of a Gaussian kernel support vector machine (SVM). Additionally, we validated the proposed methods using a large-scale transaction dataset. Motivated by our experimental results, we recently proposed three distinct quick approaches to deep learning for time-series data at the 2016 World Congress on Internet Security (WorldCIS-2016). The experimental results failed to demonstrate the effectiveness of our proposed method; however, through this work, we identified three key challenges associated with the proposed methods: (1) creating a model that generates an error rate less than 50%, (2) dividing the dataset evenly over time, and (3) considering the class not contained in the divided data. If a method can overcome these problems, then our proposed technique will be effective according to the ensemble learning theory. Herein, we apply our proposed majority rule approach to deep learning for real credit card transaction data and discuss the experimental results.

Published in: World Congress on Internet Security (WorldCIS-2017)

  • Date of Conference: 11-14 December 2017
  • DOI: 10.2053/WorldCIS.2017.0010
  • ISBN: 978-1-908320-81-0
  • Conference Location: University of Cambridge, UK