How do we check the model robustness? pls explain it for logistic regression models


#1

What is model Robustness?
Is there any specific methods for different algorithms?


#2

@ssvbalan

Model robustness can be understood as - If a model has a testing error(on a new test set) equal to the training error, then the model is said to be robust i.e the model generalises well and doesn’t overfit.

In order to explain robustness for logistic regression models, let us take an example of a binary classification problem with output having two levels- 1 and 0.

Training set - Number of 1s - 1000, Number of 0s- 500.
Test pet - Number of 1s - 500, Number of 0s- 250.

Now after training on the training set, the logistic classifier predicts all 1s as 1s and 0s as 0s perfectly but while making prediction on the test set classifies 250 1s as 0s and 200 0s as 1s.

Such a model will not be called robust since this model failed to generalise to the new dataset.

Hope this helps.