Fixing High Bias in a House-Price Regression
Problem statement
A team fits an ordinary least-squares linear regression to predict home sale prices. The training sample is representative of the market. Their diagnostics show high bias and low variance: the training error is already large, and the validation error is only slightly larger than the training error.
Which single change is most likely to reduce the bias?
Show hint
Which term of the error decomposition does the diagnosis point to, and what kind of change moves that term?
Relevant roles: ML Engineer, Applied Scientist and Data Scientist.