Various Methods for Fraud Transaction Detection in Credit Cards

Abstract

Cashless payments have become effortless with the advent of new technology and the internet. But, for online transactions, you don't have to be in a certain place where the transaction occurs, making it vulnerable to fraudulent attacks. A cyber-attacker can pretend to be the owner of a credit card and make a fraudulent transaction. There are several techniques to determine the nature of the transaction, for instance, by comparing the current transaction with previous transactions. If the monetary difference between current transaction and previous transaction is too large, then there is a greater probability of current transaction being a fraudulent transaction. This method is not reliable for anomaly detection. In some countries like India and China, banks deploy a two-step verification process which strengthens the security of the transaction. While in other countries, employees in the bank manually segregate the transactions to be fraud or not. These methods are highly dependent on human intervention. Machine Learning can be utilized to automate the process of anomaly detection. Supervised algorithms such as Logistic Regression can be used to build a model that will predict the output in the form of binary classes i.e. 0 for a valid transaction and 1 for a fraudulent transaction. Autoencoder Neural Network is one of the unsupervised algorithms using which better accuracy can be obtained for anomaly detection. In this paper, we explain different machine learning algorithms viz; Hidden Markov Model, Artificial Neural Network, and Convolutional Neural Network. Moreover, Logistic Regression is implemented, and the results obtained are highlighted.