Overfitting: Causes and Solutions (Seminar Slides)
24 Pages Posted: 26 Feb 2020 Last revised: 2 Mar 2020
Date Written: February 26, 2020
Abstract
When used incorrectly, the risk of machine learning (ML) overfitting is extremely high. However, ML counts with sophisticated methods to prevent: (a) train set overfitting, and (b) test set overfitting.
Thus, the popular belief that ML overfits is false. A more accurate statement would be that: (1) in the wrong hands, ML overfits, and (2) in the right hands, ML is more robust to overfitting than classical methods.
When it comes to modelling unstructured data, ML is the only choice. Classical statistics should be taught as a preparation for ML courses, with a focus on overfitting prevention.
Keywords: Machine learning, econometrics, backtest overfitting, selection bias, multiple testing, false discoveries
JEL Classification: G0, G1, G2, G15, G24, E44
Suggested Citation: Suggested Citation