Towards detecting credit card frauds using Hidden Markov Model
E-commerce systems have become increasingly popular due to the widespread of internet shopping and banking. Credit card is one of the mostly used forms of payment on e-commerce platforms. However, there has been a tremendous rise in fraudulent credit card transactions, resulting to huge financial losses. In this work, a Hidden Markov Model (HMM) is proposed to design a credit card fraud detection system. Each HMM specifies the likelihood of a transaction given its sequence of previous transactions. This model is driven by a combination of K-Means and Baum Welch algorithms. A clustering process, obtained by the K-Means algorithm groups each transaction based on users’ spending profiles, where each cluster is used for different hidden states of the model. Subsequently, the Baum Welch algorithm generates a trained set of observations and calculates the probability of acceptance, which is used to detect if a current transaction is fraudulent or legitimate. This approach was implemented using PHP and was tested with a simulated dataset. Four performance metrics were used on the model which includes a Fraud Detection Rate (FDR), False Alarm Rate (FAR), Accuracy (A) and Sensitivity (S). The experimental results gave a high level of FDR and a low level of FAR, indicating that the proposed Hidden Markov Model is an effective approach for detecting credit card frauds.
Keywords: Credit card, Fraud, E-commerce, Hidden Markov Model, K-Means algorithm, Baum Welch algorithm