You are reviewing a supervised learning pipeline and notice that model errors may come from both the data and the model itself. You want a structured way to diagnose whether the issue is underfitting, overfitting, or biased data collection.
How would you deal with bias in a model or dataset?