Revealing
Challenges within the Application of Machine Learning Services A
Delphi Study (pp001-029)
Robert Philipp,
Andreas Mladenow, Christine Strauss, and Alexander Voelz
doi:
https://doi.org/10.26421/JDI2.1-1
Abstracts: Over the past years,
Machine Learning has been applied to an increasing number of
problems across numerous industries. However, the steady rise in the
application of Machine Learning has not come without challenges
since companies often lack the expertise or infrastructure to build
their own Machine Learning systems. These challenges led to the
emergence of a new paradigm, called Machine Learning as a Service.
Scientific literature has mainly analyzed this topic in the context
of platform solutions that provide ready-to-use environments for
companies. We recently have developed a platform-independent
approach and labeled it Machine Learning Services. The aim of
the present study is to identify and evaluate challenges and
opportunities in the application of Machine Learning Services. To do
so, we conducted a Delphi Study with a panel of machine learning
experts. The study consisted of three rounds and was structured
according to the five steps of the Data Science Lifecycle. A
variety of challenges from the areas Communication, Environment,
Approach, Data, Retraining, Testing, Monitoring and Updating,
Model Training and Evaluation were identified. Subsequently, the
challenges revealed by the Delphi Study were compared with previous
work on Machine Learning as a Service, which resulted from a
structured literature review. The identified areas serve as possible
future research fields and give further implications for practice.
Alleviating communication issues and assessing the business IT
infrastructure prior to the machine learning project are among the
key findings of our study.
Key words:
Machine Learning as a Service, MLaaS, Machine
Learning Services, Machine Learning, Delphi study, Data Science
Lifecycle, Machine Learning Platform