The field of machine learning (ML) is of specific interest for production companies as it displays a
perspective to handle the increased complexity within their production planning and control (PPC) processes
in an economic and ecologic effective as well as efficient way. Several studies investigate applications of
ML to different use cases. However, the research field lacks in research on industry case studies. A broad
understanding from a practical perspective and in this context, an evaluation from a data mining and business
standpoint is key for gaining trust in ML solutions. Therefore, this paper gives a comprehensive overview
of evaluation dimensions and outlines the current state of research in ML-PPC by conducting a systematic
research overview. First, the present work provides key dimensions of business and data mining objectives
as evaluation metric. Business objectives are clustered into economic, ecological and social objectives and
data mining objectives are grouped into prediction accuracy, model’s explainability, model’s runtime, and
model’s energy use. Secondly, the systematic literature review identifies 45 industry case studies in MLPPC from 2010-2020. The work shows that the scientific publications only rarely reflect in detail on a wide
range of evaluation metrics. Instead, researchers mainly focus on prediction accuracy and seldom investigate
the effect of their results to a business context. Positively, some papers reflect on further aspects and can
inspire future research. This resulting transparency supports decision makers of companies in their
prioritization process when setting up a future ML-roadmap. In addition, the research gaps identified herein
invite researchers to join the research field.
|