Need Help for TWM


Need Help for TWM


i am using TWM.and i want to know is anybody knew that how to detect fraud using Teradata warehouse miner and how it can be inplemented in it? or any example anybody have or studycase for it. please share it?

thanks in advance,rgds,Dinesh Agrawal

Teradata Employee

Re: Need Help for TWM

Hi Dinesh,

For TWM, you are limited to the analytics shipped ... so for fraud you can try to make a data mining model using linear / logistic regression, decision tree mining, and / or clustering. For regression and decision trees your dependent column should be fraud/not fraud. You should have a ready-available training dataset with rows showing whether they are fraudulent or non-fraudulent transactions which are used to build the data mining model.

Your implementation in TWM will be based on creating an analytical dataset first (ADS creation modules based on variable creation and variable transformation analyses as required). The Analytical dataset (ADS) is a data-mining-ready table which meet the algorithmic requirements for data mining techniques (e.g. removing nulls, transforming columns to numerical format, aggregating values, etc). Then you should run the analytics modules to create the models (analytics modules should run on the ADS training-dataset which contain previous transactions indicating if they are fraudulent or non-fraudulent). After creating the model you can run the scoring modules which takes new "unknown" transactions/instances (i.e. new rows with unknown dependent column - in our case the fraudulent/non-fraudulent column is unknown and needs to be calculated by the data mining model you have built) and assigns a value indicating if the row (new transaction) is fraudulent or non-fraudulent.

Try to search online for "fraud-detection data mining case studies" and it will guide you examples as to what needs to be done. Then you can simply translate the steps to TWM modules as described in the previous paragraph I've mentioned.

Good luck with your endeavours!