Can Logistic Regression Reach 100% Accuracy on Overlapping Gaussians?

EasyClassical ML

Problem statement

Can logistic regression classify these training data perfectly? The classes are equally common and follow substantially overlapping normal distributions with different means and variances on a single input feature. Fit logistic regression using only that feature and assume infinitely many training samples. Explain what limits its accuracy.

Show hint

What is the lowest error that any classifier could reach on this data, and what property of the data sets it?

Relevant roles: ML Engineer, Applied Scientist and Data Scientist.

Related free problems