Integrating knowledge from DEX hierarchies into a logistic regression stacking model for predicting ski injuries

Abstract

Machine learning models are often unaware of the structure that exists between attributes. Expert models, on the other hand, provide structured knowledge that is readily available, yet not often used in machine learning. This paper proposes the integration of expert knowledge, represented in the form of multi-criteria DEX (Decision EXpert) hierarchies or attributes, in a logistic regression stacking framework. We show that integrating expert knowledge into a machine learning framework can improve the quality of models. We tested our hypothesis on the problem for predicting ski injury occurrence, an important decision-support task in ski-resort management. Our results suggest that using a DEX hierarchy of attributes and stacking improves the AUC (area under the curve) compared to logistic regression models unaware of the DEX hierarchy from 1 to 4%.