This isn't really a blog, its more of a holding page for my domain (seems a shame not to have a page), if I know you then add me on either LinkedIn or Facebook (links are on the right), however if I don't know you then I won't add you!

Wednesday 22 May 2013

Dissertation Series - Resistances to Data Mining - Accuracy


Data mining is in many cases used to forecast/predict an outcome, it does this with a degree of accuracy although it is important to note that is a forecast/prediction and not an actual.  For this reason results taken from data mining exercises should be acted on with this in mind.  Any the action taken as a result of data mining should be with the consideration as to what the impact would be if it were applied where the prediction incorrect.
These incorrect predictions are referred to as false positives, which is where something is being flagged as something it is not, where as a false negative is where something is not grouped as it should (Thrasingham, 1999 p.93).
Thuraisingham (1999 p.93) identifies the possible implications of acting on these “false positives”.
“if an agency finds incorrectly that its employee has carried out fraudulent acts and then  starts to investigate his behaviour, and if this is known to the employee, then it could damage him”
Conversely the same logic applies to false negatives “we do not want the data miner to return a result that the employee was well behaved when he is a fraud” (Thrasingham, 1999 p.93).
An area in which many consumers would have been exposed to the false positives of data mining is credit/debit card companies, as part of their fraud prevention systems banks look at consumer transaction patterns and place temporary blocks on cards that exhibit those that match patterns of stolen cards (AAAI, 2012).  A temporary block on a card requires the customer to contact their bank to unblock the card, again these false positives can be frustrating or even embarrassing for the individual concerned.
 “just because an individual makes a series of credit card purchase that are similar to those often made when a card is stolen does not mean that the card is stolen or that the individual is a criminal” (Dunham, 2003 p.16)

No comments:

Post a Comment