Offline selectors fail to beat best single model in edX dropout prediction

Arxiv CS.LG (machine learning)

A diagnostic study on edX dropout prediction finds that offline selectors trained from logged data routinely cannot outperform the best single model. The paper explores why and when selectors fail.

Arxiv CS.LG (machine learning)·Tyler CrosseAlan Nadelsticher RuvalcabaDustin Khang LeDucThomas TraskNicholas LytleDavid Joyner

7 days ago