Classification of Credit Card Fraudulent Transactions using Neural Network and Oversampling Technique

Credit card fraud is a financial type of fraud that involves the use of credit card details to purchase products and withdraw specific amounts without the permission of the person who holds the credit card. Since the advent of the online payment method in the banking sector, there has always been someone or a group of individuals who have discovered new techniques or approaches to obtaining finance/funds through unlawful means. It is noted that the card owner is not aware of illegal transactions that have performed with his/her credit card until any kind of purchase is made, as physical credit cards are not used in online purchases. In recent years, many credit card companies have deployed an automated system with machine learning technique commonly known as Fraud Detection system so as to analyse fraudulent transaction. Every new fraudulent activity raises the demand for software systems that detect fraudulent credit card transactions. Based on the logs of transactions performed, many researchers have built credit card fraud detection systems that use various data mining and deep learning techniques, machine learning algorithms to determine whether the transaction performed is fraudulent. However, the intricacy of fraudulent transactions is created in such a way that it resembles the genuine ones every time.. For the identification of frauds, the suggested method employs unbalanced severely skewed transactional data and a convolutional network. The dataset utilised here is the highly skewed machine learning kaggle dataset for credit card fraud detection. The characteristics that have been assessed are 1 for the fraud class and 0 for the non-fraud class. The present research uses credit card fraud dataset, where the dataset is preprocessed with the help of Principal Component Analysis, Adaptive Synthetic Sample technique and then Neural Network Classifiers are applied with different number of hidden layers and performance of these classifiers has been evaluated on the basis of accuracy, precision and recall rate