The Credit Card Fraud data is a small version (comprised of 12,240 records) of a much larger data set (containing 248,807 records); it is made up of 2013 European transactions. It is a very unbalanced data set in which there are only a few fraudulent transactions. Attempting to classify transactions as fraudulent will be difficult since there are very few instances of fraud.

Use Logit and a kNN model to create a predictive model for the Credit Card Fraud data. Does either of these models have predictive power? 

Explain carefully the information provided by the lift chart or the decile-wise lift chart; how does this information differ from the information provided by the overall misclassification rate? 

What value to a firm could you see in creating such a model and using it in real time?