A Simplified Perspective on Assessing Probabilistic Forecasts
This chapter introduces a simple, intuitive approach to the assessment of probabilistic inferences. The Shannon information metrics are translated to the probability domain. The translation shows that the negative logarithmic score and the geometric mean are equivalent measures of the accuracy of a probabilistic inference. The geometric mean of forecasted probabilities is thus a measure of forecast accuracy and represents the central tendency of the forecasts. The reciprocal of the geometric mean is referred to as the perplexity and defines the number of independent choices needed to resolve the uncertainty. The assessment method introduced in this chapter is intended to reduce the ‘qualitative’ perplexity relative to the potpourri of scoring rules currently used to evaluate machine learning and other probabilistic algorithms. Utilization of this assessment will provide insight into designing algorithms with reduced the ‘quantitative’ perplexity and thus improved the accuracy of probabilistic forecasts. The translation of information metrics to the probability domain is incorporating the generalized entropy functions developed Rényi and Tsallis. Both generalizations translate to the weighted generalized mean. The generalized mean of probabilistic forecasts forms a spectrum of performance metrics referred to as a Risk Profile. The arithmetic mean is used to measure the decisiveness, while the –2/3 mean is used to measure the robustness.
Oxford Scholarship Online requires a subscription or purchase to access the full text of books within the service. Public users can however freely search the site and view the abstracts and keywords for each book and chapter.
If you think you should have access to this title, please contact your librarian.