Ensemble Learning for Credit Card Fraud Detection

Timely detection of fraudulent credit card transactions is a business critical and challenging problem in Financial Industry. Specifically, we must deal with the highly skewed nature of the dataset, that is, the ratio of fraud to normal transactions is very small. In this work, we present an ensemble machine learning approach as a possible solution to this problem. Our observation is that Random Forest is more accurate in detecting normal instances, and Neural Network is for detecting fraud instances. We present an ensemble method - based on a combination of random forest and neural network - which keeps the best of both worlds, and is able to predict with high accuracy and confidence the label of a new sample. We experimentally validate our observations on real world datasets.