using clustering to check for fraud

mengkoon007

Hi,

 

I am trying to detect expense claim fraud using rapidminer. I am not too sure what is the suitable modelling technique, thus I tried out k-mean clustering. 

 

I have a huge data containing the following attributes, basically only amount is numeric and from my understanding k-mean can only use to analyze numeric.

- date

- employee

- amount

- expense type

etc

 

I have done the process and output as below: Basically, I just filter one employee at a time and select the amount attribute.

 

Qn: How can I analyze from the output to detect if there is any fraud claim?

 

Thanks.

Thomas_Ott

Fraud is always a great use case but it can be tricky to find them. Have you tried the Anomaly Detection extension? They have a great HBOS score operator. 

Telcontar120

Or if you already have some identified cases of fraud, then you can create a label and then use some of the supervised machine learning algorithms such as neural nets, random forest, or SVM.  All those are popular techniques for fraud detection (assuming you have labeled data).