Tuesday, 21 June 2011

Fraud Detection & Prevention


Financial industry is facing the fiercest competition in current time after economic meltdown. Banks are using all avenues to grow their customer base considering the survival aspect. This has lead to tremendous volume growth in banking accounts applications, credit card applications, and financial transactions. Obviously as a consequence the number of fraudulent applications and transactions is also rapidly growing.

With new payment channels like prepaid cards, e-payments & now mobile-payments, fresh opportunities for frauds are emerging.

Some of the industry research shows that:
  • Credit card frauds losses over 8 billion USD per year
  • Insurance policy holders have to pay higher premium up to 5%
  • Total fraud Losses are estimated over 30 billion USD per year
Frauds cane be classified into various categories as below:
  • Credit/Debit/Charge card fraud
  • Check fraud
  • Internet transaction / wire transfer fraud -
  • Insurance or healthcare or warranty claim fraud – over payments, false claims
  • Subscription fraud – use of telecom services with false credentials
  • Money laundering
  • Identity theft or account takeover
Analytics approaches to detect & prevent Frauds:
  • Combine historical fraud data with industry knowledge & external market data
  • Create a proof of concept to test the history data to determine fraud cases
  • If historical data is not available then anomaly detection or outlier detection is used
  • Apply the statistical model for fraud detection
  • Models are based on past spending patterns, demographic information
  • Further text mining & link analysis for probable associations to find deeper frauds
Benefits:
  • Increased number of identification of fraud cases
  • Dollar savings from fraud prevention adds to bottom line
  • Protect the customer base from financial loss or identity theft
  • Improvement in service helps to differentiate in highly competitive market
How companies are using it:
  • Financial institutions using it to identify frauds in leasing contracts
  • Banks are using it to detect credit card, wire transfers, check frauds
  • Insurers are using it to detect fraudulent claims to save the losses
  • Healthcare provider can optimize the medical loss ratio by detecting claims frauds

4 comments:

  1. This comment has been removed by the author.

    ReplyDelete
  2. This is really very good. There are few good Articles available at www.want2rich.com on Fraud Detection, Customer Segmentation and New customer acquisition techniques used in credit card Industry. Here are the Links:

    http://www.want2rich.com/2011/05/personal-finance/methods-and-tools-used-in-credit-card-industry-to-detect-fraud/

    http://www.want2rich.com/2011/05/personal-finance/market-segmentation-and-analysis-of-credit-card-industry/

    http://www.want2rich.com/2011/06/personal-finance/one-to-one-customer-intelligence-predictive-analysis/

    http://www.want2rich.com/2011/06/personal-finance/new-customer-acquisition-strategies-in-credit-card-industry/

    http://www.want2rich.com/2011/06/personal-finance/importance-of-social-media-in-credit-card-industry/

    ReplyDelete
  3. I Sandeep,
    I work in Text Mining and Data mining project in a big multinational company (reInsurance).
    To be successful in the fraud detection, the companies need to understand some key points:
    1) it is required big investments on the data quality:
    often the quality of the data is so poor that the entropy doesn't allow any proficient analysis.
    2) Apply "Divide et Impera" approach: solve small sub problems, and try to merge the single solutions: often there are some predictors of fraud that you cannot see if you don't split the business process in functional process.
    3) FORGET miraculous tools! Even for easy problem like document classification a commercial tools require so much customization that the efforts overcome the benefits!.
    4) Entrust the data analysis to serious consultants able to customize algorithms and technics for the specific domain.
    Let me bring to you attention my blog:
    There is not a specific tool for large dataset. To classify it extract a reasonable sample set and test on different test set extracted randomly!
    Of course before to do that you need to check the homogeneity of your data: (sometimes for problem having dimension >3 are useful SOM like kohonen maps).

    To be honest I don't see great problem to classify large data set. Some problem could arise when the dimension of the problem is great (in this case you could reduce the number of features with some algo like C4.5 or PCA). Yuo can find most of the algo I mentioned for example in WEKA (not useful for productive tool, but is pretty good to evaluate the benefits of the common algo).
    Let me bring to your attention my new blog site:
    http://textanddatamining.blogspot.com/

    ReplyDelete
  4. Impressive information you are shared.You are provide the one more way for detect the fraud risk.

    ReplyDelete

LinkWithin

Related Posts Plugin for WordPress, Blogger...