Credit Card Fraud Detection Using Hybrid Learning

Online banking allows a user to conduct financial transactions through the internet from any place of our convenience without physically visiting the bank. It facilitates almost all basic banking transactions and online payments efficiently and instantly. However, at the same time, internet use for money transfers is quite risky and prone to various kinds of intrusive actions. Online banking fraud is a type of malicious action or theft committed using internet technology to illegally steal money from a bank account. The distribution of attribute value is imbalanced and non-symmetric. Another challenge is that we cannot detect fraud if the user performs a low number of transactions. Due to the lack of labeled datasets, such a security analysis system must be able to work in an unsupervised or semi-supervised fashion. The main objectives of the current research are to develop an effective Banking Fraud Detection System to achieve the following points: (a) reduce skewness in the dataset, (b) accurately detect fraudulent transactions and reduce the misclassification rate. In this research, we have proposed a semi-supervised online banking fraud analysis and decision support system for online banking transactions. Tukey’s method and LSTM techniques are used to handle imbalanced fraud datasets and for classification analysis.

I. Introduction

Online/Internet banking is easy to access as well as it is prone to cyber-attacks. Cybercriminals apply several approaches to carry out fraud. Once an account holder’s account is compromised, account information can be used to steal money from the bank account. Nowadays, fraudsters are constantly improving their techniques so that fraud cannot be detected easily. Fraudulent behavior is dynamic and scatters on large and highly imbalanced datasets, so it is difficult to detect. Customer habits change over time, so it also acts as a barrier in differentiating between normal transactions and fraudulent transactions.