Financial fraud detection using machine learning

Abstract

The method of performing transactions by means of payment cards is extremely efficient and the payment card industry is rapidly growing in popularity. However, the frauds associated with the payment cards are increasing and the patterns are evolving. Although a relatively smaller percentage is detected, fraud has become a major issue that affects the global banking industry. Machine learning techniques are widely used for payment card fraud detection. The use of machine learning techniques generates successful results as there are large numbers of historical data that could be used for mining and manipulation. There are various machine learning algorithms available to construct fraud detection models. The main drawback of those models is their inability to deliver results accurately and efficiently at the level required by the industry as there is only a fine line between the fraudulent and non-fraudulent transactions. The aim of this research is to create a model that reduces the present gap in the detection of payment card frauds using the ensemble machine learning technique. Ensemble methods are learning models that achieve performance by combining the opinions of multiple weaker models. The performance evaluation of a new ensemble model has been done on the real world financial data and the results indicated its capability of identifying a high percentage of frauds with low false alarm rate than the existing models in the payment card industry. Finally, results are analyzed, interpreted and directions for further research are suggested.