Whether it be predictive maintenance, quality assurance or tumour detection – in industry and medicine, supervised learning is the most widely used machine learning method. However, the advantage of accurate predictions is offset by the considerable effort required to label training data. This task is predominantly carried out manually and, due to the large volumes of data involved, is characterised by monotony, a lack of motivation and a lack of attention.

Photo: sentin GmbH

To identify potential for optimisation in terms of human-centred design, the work process was scientifically monitored using a labelling project involving real customer data as a case study. As part of a case study, a participant labelled over 1,600 images over several working days, creating almost 4,000 annotations across 100 different classes. Before each prolonged break, the participant recorded her experiences and observations in a diary. In addition, she completed a questionnaire on her experience of flow on each occasion.

The analysis of the experience diaries and questionnaires yielded initial insights into psychological strain over time, as well as ideas for optimising software and process design. Among other things, it was found that a sense of responsibility diminished as the duration of the task increased. The findings of this study are intended to serve as the basis for a broader experimental study.

sentin, Bochum

Lehrstuhl Arbeits- Organisations- und Wirtschaftspsychologie, Ruhr-Universität Bochum

Chair of Production Systems, Ruhr University Bochum