Conformal and probabilistic prediction with applications: editorial

This special issue of Machine Learning is devoted to conformal prediction, an emerging technique in machine learning that focuses on designing prediction algorithms that are provably valid, in various senses. There have been annual meetings on this subject over the past seven years. The basis for this special issue is formed by selected papers presented in the last year’s meeting, the Sixth Symposium on Conformal and Probabilistic Prediction with Applications (COPA 2017) that took place on June 14–16, 2017, in Stockholm (with tutorials on 13 June). The authors’ of the best papers presented at COPA 2017, as judged by the Programme Committee, have been invited to submit revised and expanded versions of their papers to this special issue, and in addition, there was an open call for papers on conformal prediction and related topics. Naturally, the whole set of the submitted papers went through several rounds of peer review before they were accepted for publication. One of the advantages of conformal prediction is its flexibility; e.g., almost any known classification or regression algorithm can be turned into a conformal predictor. The simplest mode of using conformal prediction and its sister method of Venn prediction is to use it on top of traditional machine-learning techniques as a “wrapper” method that complements their predictionswith validmeasures of confidence. The underlying algorithms that have been used successfully with conformal prediction include neural networks, support vector machines, nearest neighbours methods, random forests, boosting, as well as older statistical techniques such as ridge regression and discriminant analysis. Newer ways of applying conformal and Venn prediction is to produce different kinds of predictions, such as predictive probability