4
$\begingroup$

How to compute the global precision given the precision calculated for each class? Is it just the average over classes precisions? When I use Weka the global precision is not computed as the average one, but as a "weighted average" and I don't know how this later is computed (weighted by what?). See the example bellow of the results that we get with Weka, can you please tell me how "Weighted Avg." is computed in this result (last line)?

Correctly Classified Instances         608               93.5385 %
Incorrectly Classified Instances        42                6.4615 %
Total Number of Instances              650     

=== Detailed Accuracy By Class ===

           TP Rate   FP Rate   Precision   Recall  F-Measure   ROC Area  Class
             0.889     0.005      0.727     0.889     0.8        0.995    APP05179028
             0.991     0.032      0.866     0.991     0.924      0.999    APP05179007
             0.633     0.005      0.864     0.633     0.731      0.989    APP05179012
             0         0          0         0         0          0.954    APP05179014
             0.972     0.013      0.936     0.972     0.954      0.998    APP05179010
             0.957     0.002      0.957     0.957     0.957      0.999    APP05179009
             0         0          0         0         0          ?        APP05179018
             0.6       0          1         0.6       0.75       0.988    APP05179027
             0         0          0         0         0          ?        APP05179023
             1         0.002      0.75      1         0.857      1        APP05179025
             0.8       0          1         0.8       0.889      0.991    APP05179016
             0         0          0         0         0          ?        APP05179021
             0.889     0          1         0.889     0.941      1        APP05179029
             0.918     0          1         0.918     0.957      1        APP05179020
             0         0          0         0         0          0.958    APP05179022
             1         0.002      0.963     1         0.981      0.999    APP05179015
             0.846     0          1         0.846     0.917      0.998    APP05179008
             0.992     0.002      0.992     0.992     0.992      1        APP05179011
             0.947     0.002      0.973     0.947     0.96       1        APP05179013
             1         0          1         1         1          1        APP05179026
             0.857     0          1         0.857     0.923      1        APP05179019
             0.714     0.011      0.417     0.714     0.526      0.952    APP05179017
             0.969     0          1         0.969     0.984      1        APP05179030
             1         0          1         1         1          1        APP05179024
             0.935     0.008      0.939     0.935     0.933      0.998    **Weighted Avg.**



   a   b   c   d   e   f   g   h   i   j   k   l   m   n   o   p   q   r   s   t   u   v   w   x 
   8   0   1   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   a = APP05179028
   0 110   0   0   1   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   b = APP05179007
   0   9  19   0   2   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   c = APP05179012
   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   1   0   0   1   0   0   0   0   0 |   d = APP05179014
   0   3   0   0 103   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   e = APP05179010
   0   1   0   0   0  22   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   f = APP05179009
   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   g = APP05179018
   3   0   0   0   0   0   0   6   0   1   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   h = APP05179027
   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   i = APP05179023
   0   0   0   0   0   0   0   0   0   3   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   j = APP05179025
   0   1   0   0   0   0   0   0   0   0   4   0   0   0   0   0   0   0   0   0   0   0   0   0 |   k = APP05179016
   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   l = APP05179021
   0   0   0   0   0   0   0   0   0   0   0   0  32   0   0   0   0   0   0   0   0   4   0   0 |   m = APP05179029
   0   1   0   0   0   1   0   0   0   0   0   0   0  45   0   0   0   0   0   0   0   2   0   0 |   n = APP05179020
   0   0   1   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0 |   o = APP05179022
   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0  26   0   0   0   0   0   0   0   0 |   p = APP05179015
   0   0   1   0   0   0   0   0   0   0   0   0   0   0   0   0  11   0   0   0   0   1   0   0 |   q = APP05179008
   0   0   0   0   1   0   0   0   0   0   0   0   0   0   0   0   0 126   0   0   0   0   0   0 |   r = APP05179011
   0   1   0   0   1   0   0   0   0   0   0   0   0   0   0   0   0   0  36   0   0   0   0   0 |   s = APP05179013
   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   3   0   0   0   0 |   t = APP05179026
   0   1   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   6   0   0   0 |   u = APP05179019
   0   0   0   0   2   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   5   0   0 |   v = APP05179017
   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   1   0   0   0   0  31   0 |   w = APP05179030
   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0   0  12 |   x = APP05179024
Improve this question
$\endgroup$

2 Answers 2

Reset to default
5
$\begingroup$

In principle, there is no such thing -- both precision and recall only apply to binary classification. The best you can do is to report accuracy (the per cent of correct answers) or to summarise recall/precision scores for all classes, like the precision varied from X% for CX to Y% for CY...

Any other metrics will be non-standard and will likely hinder attempts to make comparison of your results with other solutions.

Improve this answer
$\endgroup$
1
$\begingroup$

The weights are the number of instances in each class.
you can find them in weka interface after you upload the .arff file (preprocess) by clicking on the class attribute.

Improve this answer
$\endgroup$

Your Answer

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

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