Fraud Detection using Neural Network Sanjay Singh, Anchal Karnwal, Neha Prasad, Royston D’Souza, Ashwin Shenoy Department of Information & Communication Technology, Manipal Institute of Technology, Manipal, Karanataka-576104, India Email: sanjay.singh@manipal.edu Abstract Recent studies indicate that in certain people with the risk of carrying cash; plastic money has become the trend these days. Past history of transactions like card holders’ spending pattern from the previous transaction database, and other inputs like income, location, living expense etc can be compared with the current transaction details to detect credit card frauds. Deviation on transactions can indicate a fraud. There have been a few studies conducted to determine how to detect frauds by using different technology. Artificial neural networks can integrate the experts’ experience into the software so as to provide support to the banks to detect frauds. This study used Neural Network as the upcoming technology by which detection of frauds can be made easier. I. INTRODUCTION An Artificial Neural Network (ANN) is an information processing paradigm that is inspired by the way biological nervous systems, such as the brain, process information. ANNs, like people, learn by example. An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process. The simulation of artificial neural networks (ANNs) is an increasingly important research area [1] [2]. The purpose of this study is to present a simulation model using NeuroSolution for Excel (A Neural Network Software) [5] to detect frauds in credit card transactions. The network is then implemented using C++ on VC++ platform. NeuroSolution for excel is an add on to the Excel Files. The network is trained using these data. The network uses past data for training. And then it can be tested over some unseen data if the network’s learning is proper. If the network gives proper result for the testing input, then it can be used for protection to decide if the transaction is authentic. More the training data better will be the network output and thus better learning curve will be obtained. Developer has to work with NeuroSolution simulation Environment [5].The software will detect frauds in accordance with the spending patterns, expiry date, address checks and identity checks. This study will use many mathematical functions and algorithms to train the network and to get a best network output. This research supports the hypothesis that spending patterns can vary from person to person. II. METHODOLOGY For this study dummy database has been created due to unavailability of real time data from the banks, for training the network to correctly classify fraudulent transactions. Dummy data has been created based on the following fields: 1. Credit card number 2. Actual PIN 3. Entered PIN 4. Actual expiry date 5. Expiry date on the transaction details 6. Customer name 7. Merchant id 8. Terminal id 9. Amount 10. Time of transaction 11. Date 12. Reference number 13. Bank number 14. Authentication code Data has been stored in the excel sheet of NeuroSolution [5]. For making the network to learn, supervised learning technique has been used, which is used to infer the mapping implied by the data and the cost function is related to the mismatch between mapping and the data. Algorithms are used as a straight forward application of optimization theory and statistical estimation. Gradient Descent Algorithm and Least Mean Square Algorithm have been used in this study. There is use of learning algorithm to make the network learn and there is training algorithm to train the network. Learning rate (μ) is an important consideration to change the weights at each step. If μ is small it will take long time to converge and if it is very large error