Ans. There are a lot of software companies and experts claiming to do “data mining” when in fact they are performing “data reporting.” Data mining starts with data integration ,cleansing and ends with predictions with their accuracy. In addition, ranks the predictive parameters and gives users how well they account for the observed variability.

The problem with data mining is that it can’t be done well without
a) time,
b) subject knowledge
c) real data mining software(s), and
d) developing an application(DSS) for an end user.

Evaluation criteria to select a data mining tool should include:

a) Easy to Use: It should be easy to use for an expert. Data mining software should not be an MS Office look-alike, where people learning the job can intuitively play with it and get results. If they do, they’ll come to you with great graphs that point you in bizarre directions.

b) Speed: Three weeks to build a model is not something that can be implemented in today’s business. Not in data mining at least. For any type of data mining project that will actually tell you something real you may very well take a couple of months just to prepare the data.

c) Integration: All high-end software packages can integrate on all platforms and file systems. All cheaper software packages say they do.

d) Robustness: However, robustness should be a function of the model. You need to develop an appropriate model, and that is where most data mining projects fail.  

Leave a Reply