A classification system I built is going to go into production soon (it'll be part of a larger dashboard), and I'm looking for ways to better visualize and convey to business folks the results of a classification.

Basically, given "old" data on which the model was trained, I predict the classes of "new" data, with the goal being to show whether the class distributions for the new data are statistically indifferent from the class distributions observed in the old data. So, if there are three classes A, B, and C in the old data, with proportions of 50%, 30%, and 20%, respectively, I compare the classification distribution for the new data with those original, observed proportions.

Outside of a confusion matrix (which I think is probably inappropriate for most dashboard users), how else can I effectively present these results? I was thinking of a bar chart like this:

enter image description here

  • 1
    $\begingroup$ The suggestion provided by @scherm is fine if your goal is just to present the overall percentages of class predictions, but to be clear, it does not reflect the accuracy of the model. Apologies if this is out of the scope of your original question, but want to reinforce that even if your new data (predictions) has distributions 50%, 30%, 20% for classes A, B, C, it does not prove that the model is classifying properly. $\endgroup$
    – HEITZ
    Nov 22, 2016 at 23:50
  • $\begingroup$ A significant deviation from old to new could provide evidence for the new data not fitting well into the previous classes (i.e. that the predictions went horribly awry somewhere in the process). This is a problem I'm grappling with constantly, but for now I'm just trying to visualize some of these results. $\endgroup$ Nov 23, 2016 at 1:13

1 Answer 1


Here are a couple of options. I prefer the stacked column chart, which could also be annotated with the actual proportions if needed.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.