Design and Implementation of Deep Autoencoder Fraud Detection Model

The global digital growth promotes online payment transactions with plastic money card. The growth of online payment transactions attracts cybercriminals, fraudsters, and hackers to perform fraudulent online payment transactions. There are two different security measures to prevent fraudulent transactions that can be implemented including multifactor authentication and fraud detection system. Therefore, there is a need for fraud detection system to mitigate the fraudulent online payment transaction. The aim of the study is to identify and mitigate fraudulent online payment transactions. The study proposes and implements a deep autoencoder fraud detection model to detect fraudulent transactions. The training and testing of the model are performed using two datasets. The first is the primary dataset with 16 features and 880 instances and the second is the credit card fraud dataset with 31 features and 284,807, which is publicly available. Before the training and testing of the model, data anomalies from the dataset are removed using data processing and feature engineering to make the dataset compatible with the model. This study also implements a synthetic minority oversampling technique to make the dataset balanced because both datasets are imbalanced.