Evaluating Model Performance

AI Discussions
1

I have some questions regarding how to evaluate a model’s performance, and this just came to my mind:

Should I evaluate a model’s performance (such as over-fitting issues) before I add any regularization techniques or should I apply the same regularization technique to every model than cross-validate their performance?

Thanks,
Yuhan Chiang

2

Hello @Chiang_Yuhan
When it is essential to apply regularization techniques to prevent overfitting, it’s generally advisable to evaluate the model’s performance before applying regularization to understand the extent of overfitting and identify the most appropriate regularization strategy for your specific dataset and model architecture. Cross-validation can then be used to validate the performance of the regularized models and select the best-performing one.

2 Likes
3

regularisation techniques is way of fine tuning models in order to prevent overfitting.

if you do before, you will know how your model performs and then you can improvise or fine tune your model with regularization techniques to know the affects of it on your model.

If you do model performance after regularization technique, you might get either better, best or worse performance based on your parameters, dataset and /or model algorithm.

So the right answer would be you could try before and after model performance both :slight_smile: is fine and you would get a better understanding of your model as well as dataset.

Regards
DP

2 Likes