Fraud Analytics & Data Science

Fraud Analytics is considered as one of the popular domains of Data Science. With rapid advancements in the technology, financial scams are growing dramatically with far consequences in online transactions. Identifying fraudulent activities in credit card transactions has number of challenges which comprises large amount of data and high number of transactions. Data Mining had played a significant role in discovering the credit card frauds. The two risks associated with credit card fraud detection using data mining are, fraudulent patterns are changing regularly, and data is highly imbalanced as standard algorithms are failed on this synopsis. In this project, a model has been developed to detect the credit card frauds by using ensemble techniques and performing hybrid sampling on skewed data to obtain the better accuracy rate. From the experimental results the performance of meta-algorithms is compared based on efficiency, sensitivity, correctness, specificity, accuracy, Cohen’s Kappa and unbiased classification rate. Keywords- Fraud detection, meta algorithms, classification, Stacking, Bagging

The usage of credit card is increasing heavily with advancement in the e-commerce and other Internet related applications. In these days the third-party web payment gateways have become very popular in carrying out online transactions without using physical cards. Cashless-transaction system benefits influence the people to adopt online payment method which is comfortable and more convenient. However, the number of fraudsters is rising constantly with growing number of fraudulent transactions.

Credit card fraud can be classified into two types namely Inner card fraud and external card fraud. Inner card scams refer to assent occurred in between cardholder and bank by using a fake identity to commit scam whilst external fraud happens because of using cash from the pilfer card. More number of researches have been conducted on the detecting external fraud which contains major portion of credit card scams. Uncovering the fraud transactions using standard methods of traditional detection mechanisms are inefficient and ineffective. By adopting Big data analytics, the fraud detections can be identified in terms of better accuracy with less time consumption compared to manual methodologies. Data Mining is one of the significant method used to identify the fraudulent log details.

Generally, the credit card transactions data is highly imbalanced and noisy. It means only few transactions are actually fraud and majority of the transactions are genuine. By observing the transactional performance of the card holder there is a chance to predict the upcoming transaction is completed by the card holder or fraudsters. Mainly two detection techniques are used to estimate the transaction namely misuse and anomaly detections. Based on the previous transactional patterns the in-coming transaction will be predictable in anomaly method whereas in misuse detection method the it employs classification techniques to predict newly transaction is scam or not. Therefore, prediction model handling with credit card fraud detections must manage the skewed and noisy data efficiently, it should consolidate the required mechanism to endure the stream and produce valuable decisions.