Meanwhile I am working as a data analyst for an industry company. One part of my job is to research production processes and reducing defects on the products to avoid them. Unfortunately during analysis with decision trees and logistic regression models a correlation very often is given, though a causality in many cases doesn't exist. And even if a causality exists, I have to consider the direction to avoid mistaking cause and impact. All these considerations are known under the term "Business Understanding". If you don't have the business understanding for your job, data mining and more simply analytics can get misleading very fast.
Business Understanding an its spread
Nearly every company that uses analytics and data mining has a strong concern in using the insights proper, because a wrong used information can cause bad consequences. This issue is that present in the heads of many managers, that they attach a lot of importance to business understanding.
The german novel author Elias Bohst wrote an eBook ("Jagd auf Agent Albatros", in English like "Hunting for Agent Albatroaz") about a spy who wants to abscond from his job. In chapter 11 he describes a situation that shows the need of a proper business understandig by telling a story about a cat.
This cat is called Mauzi (in English spoken like "Mowtsee") and when his owner arrives at home he finds Mauzi sitting at the sill of the door near a dead bird. The human implicitly estimates the situation in a short moment and decides, that his cat has slayed the bird. Of course he is not amused and shouts at poor Mauzi, though he is not guilty. However, the estimation seems to be the closest conclusion about what happened. What the owner of Mauzi did not see were little details. Details like a spot of tallow on the glass pane of the door the bird left when he flew towards it and broke his neck. And just after this, Mauzi arrived and sat beneeth the cadaver before his owner arrived.
This story shows one big problem very well: Even the closest conclusion is not automatically the proper one. This is one of my daily problems. Very often I have to dicuss with process owners and the development department about correlations, causalities and nonsense correlations. As we found out, the closest conclusion somties just fails badly.
About the book
I've read the eBook "Jagd auf Agent Albatros" myself and I like it. The author gave me the permission to use parts of it for this post, but a condition was that I set a link to amazon.com.
Kommentare
Kommentar veröffentlichen