none
Lift Chart-Ideal line

    Question

  • I created Lift chart for model with no target value. I do not understand is there both ideal and base line like there are in lift chart with target value? Are they same?

    Also, I noticed that the lift is smaller and smaller in lift chart with target value when is ascending values on x axes and in lift chart with no target value there is different situation.


    Wednesday, September 25, 2013 10:24 PM

Answers

  • Lift chart without target value does not have random guess line, it only has ideal model line on the diagonal. Lift chart without target value is not useful in my opinion.

    Lift chart with target value is a lot more useful. In addition to lift chart for the model(s) it has lift data for the ideal model and the random guess model.

    To me, one of the most useful accuracy charts is Profit Chart. In there, I can really see what probability threshold I should use to identify rows with the target value. For example, if I am trying to identify customers that might churn, I should not use 50% as my probability threshold. I might need to do something with customers that have probability of churning> 30% or > 70% (it really depends on the cost of action per customers and how much money I am loosing on average with each churned customer).


    Tatyana Yakushev [PredixionSoftware.com]

    Download Predixion Insight 3.0 - World class predictive platform for big data

    • Marked as answer by Mary44 Friday, September 27, 2013 5:35 PM
    Friday, September 27, 2013 12:57 AM
    Answerer

All replies

  • Lift chart without target value does not have random guess line, it only has ideal model line on the diagonal. Lift chart without target value is not useful in my opinion.

    Lift chart with target value is a lot more useful. In addition to lift chart for the model(s) it has lift data for the ideal model and the random guess model.

    To me, one of the most useful accuracy charts is Profit Chart. In there, I can really see what probability threshold I should use to identify rows with the target value. For example, if I am trying to identify customers that might churn, I should not use 50% as my probability threshold. I might need to do something with customers that have probability of churning> 30% or > 70% (it really depends on the cost of action per customers and how much money I am loosing on average with each churned customer).


    Tatyana Yakushev [PredixionSoftware.com]

    Download Predixion Insight 3.0 - World class predictive platform for big data

    • Marked as answer by Mary44 Friday, September 27, 2013 5:35 PM
    Friday, September 27, 2013 12:57 AM
    Answerer
  • Dear Tatyana,

    Thank you. I have one more question, if you could help me. After execution singleton DMX query, I got results where $Support has value 993.704563. How it could be real number? This is number of cases which support the results (probability of output attribute state)?

    Friday, September 27, 2013 5:43 PM