Optimal rates for F-score binary classification

MaRDI QIDQ6318523zbMATH OpenFDO

Authors Evgenii Chzhen

Publication date 10 May 2019

Abstract: We study the minimax settings of binary classification with F-score under the $$-smoothness assumptions on the regression function $eta(x)=mathbbP(Y=1|X=x)$ for $xinmathbb{R}^d$. We propose a classification procedure which under the $alpha$-margin assumption achieves the rate $$ for the excess F-score. In this context, the Bayes optimal classifier for the F-score can be obtained by thresholding the aforementioned regression function $eta$ on some level $het{a}^{\ast }$ to be estimated. The proposed procedure is performed in a semi-supervised manner, that is, for the estimation of the regression function we use a labeled dataset of size $ninmathbbN$ and for the estimation of the optimal threshold $het{a}^{\ast }$ we use an unlabeled dataset of size $NinmathbbN$. Interestingly, the value of $NinmathbbN$ does not affect the rate of convergence, which indicates that it is "harder" to estimate the regression function $eta$ than the optimal threshold $het{a}^{\ast }$. This further implies that the binary classification with F-score behaves similarly to the standard settings of binary classification. Finally, we show that the rates achieved by the proposed procedure are optimal in the minimax sense up to a constant factor.

This page was built for publication: Optimal rates for F-score binary classification

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6318523)