Publication
Conceptual Alignment Method
Wolfgang Maaß; Arturo Castellanos; Monica Tremblay; Roman Lukyanenko; Veda C. Storey; Jonas S. Almeida
In: Twenty-ninth Americas Conference on Information Systems. Americas Conference on Information Systems (AMCIS-2023), AIS, 2023.
Abstract
This paper proposes a Conceptual Alignment (CA) Method for conceptual modeling and machine learning. The model consists of a three-step cycle that selects an initial conceptual model, aligns it with machine learning models, and refines both models to reach predictive consistency. Alignment is based on composition methods that can be instantiated by methods that satisfy contribution properties. The Conceptual Alignment Method is applied to a healthcare use case on hospital inpatient discharges. The machine learning model trained for total cost predictions is aligned with a conceptual model. We show how this refined conceptual model is used for explaining the machine learning model for a very large healthcare dataset.