Improving Fraud Detection in Credit Card Transactions Using Autoencoders and Deep Neural Networks
Open AccessIn recent years, the use of credit cards has increased significantly due to digitization and the emergence of cashless transactions. There has been a huge jump in fraud in credit card transactions across banks, credit unions, and other financial institutions across the globe. Financial institutions and credit card companies are now required to detect credit card frauds in real time to prevent further losses.This research proposed a method to improve fraud detection using unsupervised deep learning classifier autoencoders and deep neural networks. It is difficult to train machine learning models for credit card fraud detection because of the class imbalance. To build effective classifier or predictive model it is necessary to balance the number of legitimate and fraudulent transactions, data resampling was applied to the dataset through random undersampling and random oversampling, specifically the synthetic minority oversampling technique (SMOTE) and adaptive synthetic (ADASYN). Since they introduced noise in the data during resampling process, an unsupervised deep learning–based autoencoder algorithm was used for denoising. Using the denoised dataset, a deep neural network was built to classify transactions in the sampled dataset as normal or fraudulent. To implement autoencoder, this research used Google’s TensorFlow library, which is part of the Tensorflow framework. Model performance was evaluated using metrics such as precision, recall, F1 score, and area under the curve–receiver operating characteristic. The study model outperformed existing models in the industry.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.