Subscribe to DSC Newsletter

Hi,

Can anyone tell me in general how does Fraud Analytics wrok. By this I mean how do you go ahead with any kind of fraud analytics

Views: 187

Comment

You need to be a member of AnalyticBridge to add comments!

Join AnalyticBridge

Comment by SHOUNAK GHOSAL on June 14, 2010 at 4:31am
Hi Jai,

You are right that predictive analytics can be used to identify fraud. But few things we need to keep in mind while developing this kind of model.

First: - When do you want to apply your model in entire customer / claims life cycle? Be it a credit card fraud or insurance fraud, predictive model is applied to define the filtering criteria at very first stage of entire process. This helps the business to identify fraud even before it happens. So, variable selection is an important step for you. You need to have a clear business knowledge and also well defined data dictionary with you while developing the model. Once you develop the model, you will be able to tell the SIU team about list of suspicious cases with varied range of score associated with it.

Second: - It may happen that your model is asking SIU team to investigate a large number of cases to identify final fraud. Is the business ready to incur that much of cost for investigation? If not, you need to apply some other logic of clinostat or association rule (Social networking analysis) as a top up of your predictive model to reduce your previous list.

Third: - Once you finalize your entire analysis apply it on current data set and check if your list of suspicious cases provides right guidance to SIU team in-terms of identifying actual fraudulent activities.

Be careful, you may think of identify fraud only but business will take a call on the basis of opportunity cost associated with fraud leakage and customer satisfaction.

All the best for your project.

Thanks,

Shounak Ghosal
Comment by Jai Shanker Singh on June 11, 2010 at 8:48am
Thanks everyone for replying,

One thing is clear that predictive analytics can be used to model some frauds (e.g. credit card frauds etc.). Then next question that comes is if predictive analytics is not the only approach then which are the other approaches to model fraud and how how to go about it.

I am asking this because fraudulent behavior is increasing and one need to know how to predict it before it happens
Comment by Larry on June 3, 2010 at 11:52am
Fraud Analytics can be considered a segment of predictive analytics. The goal is to try to predict fraud behavior from given known data of a particular organizational process. Such fraud behavior predictions could be defaults on loans or purchasing with stolen credit cards. The idea is that you try to identify key attributes that will likely predict that bad behavior.

There are many statistical and analytic methods for fraud analytics. There are logistic regression, neural networks, decision trees, random forests, machine learning, data mining, business rules and so on. There are a lot of factors that determine which method is used to apply to a given problem.

Here is a decent read on the subject http://en.wikipedia.org/wiki/Predictive_analytics

On Data Science Central

© 2019   AnalyticBridge.com is a subsidiary and dedicated channel of Data Science Central LLC   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service