Simple Graphs for Complex Prediction Functions

Abstract

By supervised learning with p predictors, we frequently obtain a prediction func-tion of the formy = f(x1;:;x p). When p 3, it is not easy to understand theinner structure off, except for the case the function is formulated as additive. Inthis study, we propose to usep simple graphs for visual understanding of com-plex prediction functions produced by several supervised learning engines such asLOESS, neural networks, support vector machines and random forests.