From Idea to Implementation: Real-Time & Real-World Classification Testing in Nursing Practice Research

In recent years, the shortage of nurses has led to increased responsibilities and workloads. Automated nursing documentation is a viable way to reduce the workload of nurses, reduce overtime, and improve the quality of documentation. In previous work, we have evaluated the performance of classifying nursing activities using different machine learning algorithms through offline experiments, achieving accuracies of up to 95.61 %. While offline evaluations provide an initial understanding of model performance, real-time testing and classification under real-world conditions provide a more comprehensive assessment of practical applicability. In this paper, our approach is applied in a scenario involving live classification and real-world conditions. In order to automate the documentation tasks of carers, smartwatches are worn by carers which classify their care activities in real time. To validate the approach, the study focuses on three complex care activities, divided into six sub-activities, which are evaluated in real-time testing. Previous research has shown that Long Short-Term Memory (LSTM) networks produce promising results in Human Activity Recognition (HAR). Consequently, an LSTM model was used for this task. Ultimately, the LSTM model achieved a peak accuracy of 82.63 % in offline testing and 73.29 % in live testing. These results confirm the feasibility of using HAR and DL to automate nursing documentation tasks. The live evaluation phase further confirmed the practicality and robustness of the approach. In addition, this approach allows care activities to be evaluated under real conditions by filtering out undefined movements, which significantly improves live evaluation. Furthermore, the results suggest that HAR in the context of nursing is a viable solution with the potential to significantly support nurses in the long term. However, challenges remain, including the need to understand nurses’ perspectives on the technology and to evaluate a wider range of tasks.