An Ameliorated method for Fraud Detection using Complex Generative Model: Variational Autoencoder

Abstract

Perpetrating fraud for financial gain is a known phenomenon, in this fast-growing adoption of smart phones and increased internet penetration, embracing digital technology. Evolution of financial transactions over the years, from paper currency to electronic media, leading the way in the form of credit cards or interbank electronic transactions. Consumers trending towards e-commerce hasn't deterred criminals, but considered this as the opportunity to make money through defrauding methods. Criminals are rapidly improving their fraud abilities. The current Supervised and Unsupervised Machine Learning Algorithm approaches to the discovery of fraud are their inability to learn and explore all possible information representation. The proposed system, VAE based fraud detection, which uses a variational autoencoder for predicting and detecting of fraud detection. The VAE based fraud detection model consists of three major layers, an encoder, a decoder and a fraud detector element. The VAE-based fraud detection model is capable of learning latent variable probabilistic models by optimizing the average value of the information observed. The fraud detector uses the latent representations obtained from the variational autoencoder to classify whether transactions are fraud or not. The model is applied on real time credit card fraud dataset. The experimental results show that, implemented model perform better than supervised Logistic Regression, unsupervised Autoencoders or Random Forest ensemble model.