AnalyticBridge

A Data Science Central Community

Credit card Fraud Detection

Hey All,

I am working on my university project, The project is about detecting fraudulent transactions.

The data that I have is labelled (fraudulent transactions or legitimate), According to my research so far I have come up with a scheme that I will be following:

- Make clusters on the dataset without any fraudulent transactions (based on amount spent, frequency of transactions, modes, channels etc).

- Once the clusters are created, either NN or BN will be used to identify the cluster of new incoming transaction.

My question is, am I doing it the right way? What kind of clustering algorithms will be best suited? I am making clusters like :

cluster 1:

+ transaction amount: high

+ channel used: ATM

+ POS Entry Mode: EMV, Magnetic stripe, EMV fallback, cardless etc

few other rules

cluster 2:

+ transaction amount: high

+ channel used: Internet

+ POS Entry Mode: CVV, VBV, dynamic passcode etc

few other rules

Your suggestions in this regard will be highly appreciated.

Thanks & Regards,

Ali

Tags: Network, Neural, card, clustering, credit, detection, fraud, prediction

0 members like this

▶ Reply to This

Replies to This Discussion

Permalink Reply by ANKUR JAIN on July 19, 2011 at 2:57am

Hi Ali,

You have not mentioned anywhere what is your total data size and do you have confirmed fraud cases in that data set. And if yes then in what propotion they are.

You want to move further with supervised or non- supervised learning technique.

Regards,

Ankur Jain

▶ Reply

Permalink Reply by Rahil on July 19, 2011 at 9:10am

Hi Ankur,

Thanks for the reply, my data consists of 45000 records out of which 5% is fraudulent transactions.

I want to go with un-supervised approach, I want to know if the idea that I have mentioned in my first post make sense?

Thanks

▶ Reply

Permalink Reply by ANKUR JAIN on July 19, 2011 at 11:00am

Hi Ali,

In unsupervised learning technique model will be trained to predict the record which shows unusual pattern.

There are different tecniques to build model for this.

you can start with your approach but need to check with validation data set whether your model is well trained to predict the abnormal cases or not.

AnalyticBridge

Credit card Fraud Detection

Replies to This Discussion

Resources

On Data Science Central