Thesis project: Three explanatory models for business dashboards: an in-depth evaluation

This thesis entails the subject of explanatory analytics for business dashboards and addresses a shortcoming of business dashboards. In general, business dashboards provide an overview to relevant management information. However, business dashboards lack explanatory capabilities. An in-depth evaluation provides comprehensive knowledge on three models for explanatory analytics: Explanation Formalism, Informative Summarization, and Explanation by Intervention. An extensive and systematic literature review summarizes the previously scattered knowledge on the three technical models for explanation. Each of the models is studied on components, algorithm(s), and filtering capabilities. Multiple factors, amongst which the direction of recursion, data assumptions, and completeness of explanation are described and it is argued which model is best in handling hidden causes. One of the models, Explanation Formalism, is then extended through the development of a Python module capable of automatically transforming a star-structured database to a system of additive drill-down equations, in which a data point on the left-hand side of the equation is expressed in an additive function on the right-hand side of the equation. This system of additive drill-down equations is a requirement for the implementation of the Explanation Formalism method.

In addition, an operational comparison between the models is included. The models of Explanation Formalism and Informative Summarization are invoked on the same use case, and it is studied how the result of explanation differs. Resulting from this study is the knowledge that Explanation Formalism is the most comprehensive and mature method of generating explanations. Each of the models has its strengths, however, Explanation Formalism is most likely to find its way to commercial use due to the fact that the explanations can be visualized and integrated into business dashboards.

Keywords: Explanatory Analytics, Diagnostic Analytics, Automated Business Diagnosis, Business Intelligence.

Marko Lubina (markovolubina@gmail.com) is the author of this thesis.

Leave a Reply

Your email address will not be published. Required fields are marked *