suggestion - roc curve data points

Hi, i am using ROC to determine the best model but after choosing the best one i try to find best treshold for model probability to give me needed outcome - right proportion between accuracy rates - to do this i need to make a simple loop for calculating the outcomes for given probability levels as ROC node does.

is it possible to create one more output for roc node with generates data point which are then plotted on ROC image? that would be useful for further optimisation

Hi @zarniak

interesting idea. Can you share the loop you are doing here?

Did you see our component for this?

Best, Iris

2 Likes

If you want to optimize for accuracy, have a look at Palladian’s “Threshold Analyzer” node:

It’ll draw a plot of Precision, Recall, F1 and Accuracy (though Accuracy is currently not mentioned in the docs – note to myself) for an interval of thresholds from 0 … 1. Beside that, it’ll give you an output table with all the values from the graph – you can use this to find the optimal threshold.

4 Likes

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.