| How do you usually deploy data mining models (up to 4 choices), 95 voters | |
| Don't deploy, but use results to gain business or scientific knowledge (31) |  32.6% |
| Don't deploy, but publish research papers (22) |  23.2% |
| -----deployment options---- | |
| Deploy using a data mining tool for scoring (34) |  35.8% |
| Deploy using a DBMS and SQL (24) |  25.3% |
| Deploy using Java (16) |  16.8% |
| Deploy using another method(15) |  15.8% |
| Deploy using PMML (13) |  13.7% |
| Deploy using a cloud (11) |  11.6% |
Comments
Gregory Piatetsky-Shapiro, Deployment increases, especially using PMML and CloudAlthough the KDnuggets 2006 Data Mining Deployment Poll, asked a very similar question, it did not have a limit on number of choices, while 2009 poll only allowed up to 4 options. As a result we see fewer total choices in 2009 poll so it may be a better reflection of the more common deployment methods.
With these caveats, note that percent of people who selected "do not deploy" options ("publish" and "use results to gain knowledge") has declined. Examination of results shows that some of them actually selected deployment options as well. Among the deployment options, we see
- Deploy using a data mining tool for scoring - declined from 49.5% to 35.8% of respondents
- Deploy using a DBMS and SQL - increased from 21% to 25.3%
- Deploy using Java - unchanged at 16.8%
- Deploy using PMML - increased from 4.2% to 13.7%
- Deploy using a cloud - new question in 2009 - 11.6%
Will Dwinnell, Deployment Platform: Other
I deploy some predictive models to a an industry-specific vertical application.