Estimating Multi-label Accuracy using Labelset Distributions

TL;DR

This paper proposes methods to estimate confidence in multi-label classifiers by deriving expected accuracy from labelset distributions, improving interpretability and decision-making in real-world applications.

Contribution

It introduces and evaluates candidate functions to estimate expected accuracy from probabilistic multi-label distributions, enhancing confidence calibration.

Findings

Three candidate functions correlate well with expected accuracy.

Combining candidates improves estimation for Jaccard and exact match.

Methods are robust across different multi-label accuracy metrics.

Abstract

A multi-label classifier estimates the binary label state (relevant vs irrelevant) for each of a set of concept labels, for any given instance. Probabilistic multi-label classifiers provide a predictive posterior distribution over all possible labelset combinations of such label states (the powerset of labels) from which we can provide the best estimate, simply by selecting the labelset corresponding to the largest expected accuracy, over that distribution. For example, in maximizing exact match accuracy, we provide the mode of the distribution. But how does this relate to the confidence we may have in such an estimate? Confidence is an important element of real-world applications of multi-label classifiers (as in machine learning in general) and is an important ingredient in explainability and interpretability. However, it is not obvious how to provide confidence in the multi-label context and relating to a particular accuracy metric, and nor is it clear how to provide a confidence which correlates well with the expected accuracy, which would be most valuable in real-world decision making. In this article we estimate the expected accuracy as a surrogate for confidence, for a given accuracy metric. We hypothesise that the expected accuracy can be estimated from the multi-label predictive distribution. We examine seven candidate functions for their ability to estimate expected accuracy from the predictive distribution. We found three of these to correlate to expected accuracy and are robust. Further, we determined that each candidate function can be used separately to estimate Hamming similarity, but a combination of the candidates was best for expected Jaccard index and exact match.