Skip to main content

Use error analysis before trying to improve your model

When faced with poor model performance, most machine learning practitioners will spend time tuning hyperparameters or testing different model architectures. There is no need to panic or think that there is an issue with your model when you first see low metrics. Take some time to look into the root cause of performance issues before you try to change any part of your model. If you are on LandingLens, the platform will automatically select good model settings so you will see more performance gains from iterating on your data first. Walk through this approach using the steel defect dataset from when you were developing a defect book.

This section will go over best practices for this stage of the ML lifecycle:

error_lifecycle