As mentioned in our previous blog entry, appropriate and simple pictorial representations (e.g., charts etc.) can help limit the information overload that typically arises when examining results in Data Mining. Now, here are some lessons we have learned when analyzing data:
First, as in the case of Data Mining itself, the results when finding patterns etc. need to be charted in domain-specific and context-sensitive ways. That is, what may work in the field of Competitive Marketing for vehicle manufacturers would not necessarily work well for Molecular Biology research findings. And yet we often see a “one size fits all” approach of trying to use visualizations developed for one area being foisted on some entirely different effort.
Second, simple and clean charts go a long way in easily assimilating useful information in a compact manner. While snazzy, very 21st-century technologies, may be pleasing to the eye – care should be taken to assess whether the visualizations actually help in any significant way. In fact, the underlying results should lend themselves to analytical computations, whereas the visualizations are only a simpler means to make it easier for us humans to assimilate the findings.
Third, there are many charting, reporting and visualization software products (i.e., these are a commodity technology). Most of such software can be easily utilized with specialized analytical tools. And so, decisions in obtaining software should be guided by the analytical aspects needed, rather than the simpler charting or reporting capabilities (which can be easily integrated – click the thumbnail image at the right to view our screencast on Ubiquiti reporting ). Also, beware of “all-in-one” software (i.e., “general purpose” toolsets) which tend not to work too well without immense customizations (e.g., see our previous posts, “Text Mining” – A Misnomer, and Minding Data Mining).