Waruna, we do not know a data set yet, if there is one we would love to use
it.

One thought is taking KDD network activity dataset ( which is well known
benchmark for anomaly detection) and create a new dataset using that.

--Srinath



On Tue, Nov 18, 2014 at 1:48 PM, Seshika Fernando <[email protected]> wrote:

> Well the first usecase we are aiming for is Credit Card Fraud Detection at
> Merchant level. We do not have readily available datasets (since these are
> credit card transactions, usually this data is not readily available).
> However, at this point we are only doing a POC to show that our system is
> able to perform this kind of functions. It is only a matter of fine tuning
> the weights assigned to the rules, which can be done when we implement this
> for customers.
> The weights are largely dependent on the merchant type/size etc;
>
>
>
> On Tue, Nov 18, 2014 at 1:37 PM, Waruna Perera <[email protected]> wrote:
>
>> Hi Seshika,
>>
>> Are you planning to test this based on some real world scenarios and
>> data? ( ex: stock prices) Therefore to get and idea how accurate we can be.
>>
>> Thanks,
>>
>> On Tue, Nov 18, 2014 at 10:59 AM, Seshika Fernando <[email protected]>
>> wrote:
>>
>>> Hi all,
>>>
>>> Following the implementation of Fraud Rules, and Markov Chain capability
>>> in order to do outlier detection in CEP, we are hoping to implement Fraud
>>> Scoring capability.
>>>
>>> Fraud Scoring is a mechanism to evaluate multiple features of a
>>> transaction (eg:- geolocation, ip address, billing/shipping address,
>>> transaction velocity etc;) and based on historical trends and blacklists,
>>> compute a score for each transaction (eg:- between 0 and 100). Higher the
>>> score, higher the risk that the transaction will be a fraudulent
>>> transaction. [1] is a good introduction to Fraud Scoring.
>>>
>>> Now that we have already implemented several fraud detection rules, the
>>> plan is to augment this using a scoring system, so that siddhi calculates a
>>> score for each transaction,  based on how it performed in the rules. [2]
>>> gives a very simple example of how this might be done.
>>>
>>> When we complete this, we are able to show that CEP can perform fraud
>>> detection in the following ways
>>> a. Rule based
>>> b. Using Markov Chains
>>> c. Using Fraud Scores
>>>
>>> 1. http://www.fraudpractice.com/fl-fraudscore.html
>>> 2. https://www.maxmind.com/en/ccfd_formula
>>>
>>> Cheers,
>>> Seshika
>>>
>>>
>>>
>>> _______________________________________________
>>> Architecture mailing list
>>> [email protected]
>>> https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
>>>
>>>
>>
>>
>> --
>> Waruna Perera
>> Senior Software Engineer - Test Automation
>> Mobile: +94 77 3867037
>> WSO2, Inc.; http://wso2.com/
>> lean . enterprise . middlewear.
>>
>> _______________________________________________
>> Architecture mailing list
>> [email protected]
>> https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
>>
>>
>
> _______________________________________________
> Architecture mailing list
> [email protected]
> https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
>
>


-- 
============================
Srinath Perera, Ph.D.
   http://people.apache.org/~hemapani/
   http://srinathsview.blogspot.com/
_______________________________________________
Architecture mailing list
[email protected]
https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture

Reply via email to