Empirically Estimable Classification Bounds Based on a New Divergence Measure

TL;DR

This paper introduces a new divergence measure that improves bounds on binary classification error rates, applicable to both matched and mismatched training-test data scenarios, validated through feature selection algorithms on speech tasks.

Contribution

It proposes a non-parametric f-divergence measure to derive tighter classification bounds and demonstrates its practical utility in feature selection for speech classification.

Findings

Improved bounds on classification error rates.

Effective feature selection algorithms based on the new divergence.

Validated results on speech classification tasks.

Abstract

Information divergence functions play a critical role in statistics and information theory. In this paper we show that a non-parametric f-divergence measure can be used to provide improved bounds on the minimum binary classification probability of error for the case when the training and test data are drawn from the same distribution and for the case where there exists some mismatch between training and test distributions. We confirm the theoretical results by designing feature selection algorithms using the criteria from these bounds and by evaluating the algorithms on a series of pathological speech classification tasks.