Masters Thesis

Credit card fraud detection in real-time

With the rapid growth of the economy in the last two decades, our process of executing transactions has completely transformed. The online banking systems and credit card companies together process tons of transactions every day. Although with the ease in usability comes the threat of fraud transactions where cybercriminals exploit the loopholes for their benefits resulting in loss to the user or the service provider. These losses may range anywhere between a hundred to a few hundred dollars in every individual transaction but all together they pile up in huge losses to the banking organizations. The losses incurred to fraud as of 2018 have piled up to $9.47 billion just for the United States itself. To prevent these losses we have proposed a real- time fraud detection system embedded with a machine learning model with Artificial Neural Networks (ANN) and further using apache Airflow which completely automates the job and updates the Fraud Manager with emails with an added alert functionality. The model is trained and validated with the help of data of a user history of credit card fraud transactions. Since the dataset is highly unbalanced it is first undersampled and split into an 80:20 ratio for training and validation respectively for the ANN model. ANN simply provided us with the highest accuracy and speed when compared to other machine learning techniques since the structure and working of ANN is very closely similar to a human brain.