HR - Talent forecasting

Chris_AxeChris_Axe Member Posts: 1 Contributor I
Hello Community!

Straight to subject if you don't mind :)

My company need to know witch persons from HR database is qualifying for Talent pool. I have tried to find some basic rules to write a PL/SQL procedure, but it seems the process is much heuristic that i thought! I will transpose that in next simple fact: weight of attributes changes dynamically for each person and i should adapt code permanently and that it's not a viable solution i guess :). After comparing my results with an HR expert pool list, recognition of talents with my script is bellow 50%...
After some diggings on the web i get to next conclusion: Certainly RapidMiner could help me to get much better results! My expectations is around 75% of correct identifications.
And now some data:
-number of employees around 8000
-number of employee attributes 9

What do you think? It can be done? The dataset is too small to get expected results?

kind regards.


  • Options
    IngoRMIngoRM Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Community Manager, RMResearcher, Member, University Professor Posts: 1,751 RM Founder

    What do you think? It can be done? The dataset is too small to get expected results?
    Straight to the answer if you don't mind: Cannot say it  ;)

    The size of the data set is not important (although 8000 samples for 9 attributes usually sounds good to me). The problem is that I don't have any idea what is encoded by those 9 attributes. If it is merely stuff like Name, Address, Zip Code etc. you will probably end up with bad results. If they are really describing some skills etc. this might work out well for a well-designed and optimized process. If you reach 51%, the desired 75% or 99,99% will be evaluated during model creation and optimization and usually is not something which can be guaranteed before - at least not by somebody not familiar with the details and hardly any experience on this specific topic...

    Just model the data and evaluate it - then you will know!

Sign In or Register to comment.