Deep and interpretable regression models for ordinal outcomes

Outcomes with a natural order commonly occur in prediction problems and often the available input data are mixture of complex like images tabular predictors. Deep Learning (DL) models state-of-the-art for image classification tasks but frequently treat ordinal outcomes as unordered lack interpretability. In contrast, classical regression consider outcome’s yield interpretable predictor effects limited to data. We present neural network transformation (ontrams), which unite DL approaches. ontrams special case trade off flexibility interpretability by additively decomposing function into terms using jointly trained networks. The performance most flexible ontram is definition equivalent standard multi-class model cross-entropy while being faster training when facing outcomes. Lastly, we discuss how interpret components both on two publicly datasets.