Calculate classification probabilities for more than two classes

In SPSS Statistics and in open source languages (e.g. R), when you use a classification algorithm to classify more than two groups (e.g. 4 groups), the probabilities are provided for each class (not just the predicted class). This is important because it allows me to construct ROC charts and calculate the AUC for each class. I need these metrics to assess performance and also for presentations. If a sample is incorrectly classified the probabilities also allow me to assess the likelihood that a sample belongs to the another classes (individually).

For example, I may use modeler to classify an unknown rock type to one of four rock types, using one of the many great algorithms available in SPSS Modeler. Before deplying this model I have already constructed a cross validated model as an SPSS Modeler stream, but where samples are misclassified I cannot assess the probability of the sample being classified as one of the other 3 groups - this is important. I also cannot use the results in Modeler, or in R, to generate ROC charts or calculate the area under the curve. I need the ROC chart and AUC for company reports. I am OK with making the ROC charts and calculating the AUC outside of modeler, but I cannot do this without getting the probabilities from Modeler.

Unfortunately in my field many of our classification problems are not binary.

Needed By

Yesterday (Let's go already!)

Post comment

Admin

Gene Azad

Reply
| Feb 16, 2022

The team is currently taking a look at this.

0 reply Hide replies

By clicking the "Post Comment" or "Submit Idea" button, you are agreeing to the IBM Ideas Portal Terms of Use.
Do not place IBM confidential, company confidential, or personal information into any field.

Please enter your email address

RELATED IDEAS

Calculate classification probabilities for more than two classes