Frauddetection Analytics Using Decision Rules

For Financial sector, Fraud detection is the most important exercise in order to identify fraud transactions at ATM and other channels

. This will greatly help in reducing customer distress and identifying loopholes in the system further helping in removing the same.

In Fraud detection analytics, basically you need to define rules which will help us to identify whether it is fraud transaction or normal one. Basically in terms of statistics it is like whether a new entry or transaction is an outlier if I compare it with the existing cases of non-fraud cases distribution. Once you identify whether a particular transaction is a fraud one then necessary step can be taken to avoid the same.

To deep dive into the technical intricacies of the analytical technique, generally the population of the abnormal cases are very low compared to the normal cases. So applying probit regression analysis to will not give good results. Generally probabilistic regression techniques are better when you have comparable cases of both sets. So in order to avoid the above stated problem we will tell you other technique.

In this technique we take set of normal cases and develop a model on this set. We apply advanced optimization techniques to fit the data and derive a probabilistic model which fits the data. Now we can use the same model to test the anomalous cases. Suppose if the probability comes below some threshold value than we can say that it is not fitting in the data this it is an outlier or anomalous case.

Generally financial sector uses decision analysis to identify the fraud cases. In this we define a set of variables which we think are relevant in predicting whether it is a normal or anomalous case. Than we try to fit in the decision rules for different variables and identify the proportion of the anomalous cases which has been accounted by the criterion. Like this you keep on defining threshold for various variables, unless you identify a satisfactory proportion of frauds in the criterion.

For both the techniques it is very crucial to identify the relevant variables. Because only relevant variables for fraud detection will differentiate the normal cases from the fraud ones, more difference in probability density distributions between anomalous and normal cases will help you better to draw a decision boundary and segregate the cases.

The problem with the decision rules is that it takes into account both normal and anomalous cases and then builds the model. But the problem is as the population of anomalous cases is very less; you cant generate rules for bad cases. Rather the other technique builds a model based on normal examples, thus in case new anomaly comes it may take that into account.

As of now, financial sector mostly uses decision rules to identify the frauds and it works pretty well in this domain. But with time the sector will require to develop better analytical techniques in order to produce better results. Moreover, as of now banks use this as a post analysis exercise but there should be some technology integration in the same which will help to stop the transaction in case system predicts it as fraud in real time.

by: Business: Marketing and Advertising: Market Research

POST COMMENT

4 Tip To Keepyour Dog On The Right Weight Att Aldrig Mer Flytta Eller Kanske Bara En Gng Till Clotted Cream Can Add Richness To Many Dishes Botox For Cervical Dystonia 6 Consejos Esenciales Para La Integracin De Pinterest Para Bloggers How To Conserve Space In A Crowded Kitchen Maicon Injured The Truth About Six Pack Abs Review Why You Need To Declutter Your Closet This Weekend Different Smells Of Wood How To Get Better Gas Mileage By Using Some Of The Gas Mileage Devices Available Today Property Buyers Delayed By Hsbcs Panel Of Solicitors How Restaurants Can Benefit From A Pinterest Account