In a domain with as little signal as finance, knowing which features actually carry predictive information is not a nice-to-have, it is the difference between a real strategy and an overfit one. Feature importance is how you find out, and it is a staple of quant interviews because the two standard methods both have a sharp, finance-specific failure: correlated features. This lesson explains the two workhorses, mean decrease impurity (MDI) and mean decrease accuracy (MDA), works each one through with numbers, and then spends real time on the substitution effect, the way correlated features mislead MDI by diluting importance and mislead MDA by masking it. It assumes the tree models from Module 2 and the purged cross-validation from earlier in this module, and it sets up the backtest-overfitting lesson that closes the course.
Table of Contents
Already have an account? Log in!