Learning temporal representation of transaction amount for fraudulent transaction recognition using CNN, Stacked LSTM, and CNN-LSTM

摘要

This paper aims to explore deep learning model to learn short-term and long-term patterns from imbalanced input dataset. Data for this study are imbalanced card transactions from an Indonesia bank in period 2016-2017 with binary labels (nonfraud or fraud). From 50 features of the dataset, 30 principal components of data contribute to 87 % of the cumulative Eigenvalues. This study explores the effect of nonfraud to fraud sample ratio from 1 to 4 and three models: Convolutional Neural Network (CNN), Stacked Long Short-term Memory (SLSTM), and Hybrid of CNN-LSTM. Using Area Under the ROC Curve (AUC) as model performance, CNN achieved the highest AUC for R=1,2,3,4 followed by SLSTM and CNN-LSTM.