The PAV algorithm optimizes binary proper scoring rules
This work provides a theoretical extension for calibration methods in machine learning, which is incremental but improves robustness for binary classification tasks.
The paper tackled the problem of calibrating probabilistic outputs from binary pattern classifiers by showing that the pool-adjacent-violators (PAV) algorithm optimizes calibration for all regular binary proper scoring rules, extending previous results limited to convex rules, and also applies to log-likelihood-ratio calibration independently of prior probabilities.
There has been much recent interest in application of the pool-adjacent-violators (PAV) algorithm for the purpose of calibrating the probabilistic outputs of automatic pattern recognition and machine learning algorithms. Special cost functions, known as proper scoring rules form natural objective functions to judge the goodness of such calibration. We show that for binary pattern classifiers, the non-parametric optimization of calibration, subject to a monotonicity constraint, can be solved by PAV and that this solution is optimal for all regular binary proper scoring rules. This extends previous results which were limited to convex binary proper scoring rules. We further show that this result holds not only for calibration of probabilities, but also for calibration of log-likelihood-ratios, in which case optimality holds independently of the prior probabilities of the pattern classes.