Sara Santos

Office workers are exposed to a range of occupational health risks, including prolonged sedentary behaviour, postural load, elevated heart rate, and noise, yet objective and continuous monitoring of these risk factors in workplace settings remains uncommon. This study aimed to co-design occupational health visualisations based on smartphone and smartwatch data, through a multi-stakeholder group of office workers and occupational health professionals. A generative co-design framework was applied, comprising a pre-design phase with a field study and questionnaire, a structured multi-stakeholder workshop, and a follow-up evaluation session. Thematic analysis of the workshop transcript yielded 17 occupational health themes, which were subsequently assessed for technical feasibility relative to the available sensing platform. Of the 27 discrete visualisation elements proposed across both groups, the majority were classified as directly addressable using smartphone and smartwatch sensor data. Visualisations covering physical activity, heart rate, environmental noise exposure, and postural load were implemented in Python using real-world data collected from office workers. The follow-up session provided qualitative confirmation that the developed visualisations were interpretable and aligned with the stakeholder expectations. The generative co-design framework proved well-suited to the occupational health visualisation context, enabling structured translation of stakeholder requirements into technically feasible and interpretable visualisation outputs.

Office workers are exposed to a range of occupational health risks, including prolonged sedentary behaviour, postural load, elevated heart rate, and noise, yet objective and continuous monitoring of these risk factors in workplace settings remains uncommon. This study aimed to co-design occupational health visualizations based on smartphone and smartwatch data, through a multi-stakeholder group of office workers and occupational health professionals. A generative co-design framework was applied, comprising a pre-design phase with a field study and questionnaire, a structured multi-stakeholder workshop, and a follow-up evaluation session. Thematic analysis of the workshop transcript yielded 17 occupational health themes, which were subsequently assessed for technical feasibility relative to the available sensing platform. Of the 27 discrete visualization elements proposed across both groups, the majority were classified as directly addressable using smartphone and smartwatch sensor data. Visualizations covering physical activity and sedentary behaviour, heart rate, environmental noise exposure, and postural load were implemented in Python using real-world data collected from office workers. The follow-up session confirmed the interpretability and clarity of the developed visualizations. The generative co-design framework proved well-suited to the occupational health visualization context, enabling structured translation of stakeholder requirements into technically feasible and interpretable visualization outputs.

This article presents PrevOccupAI-HAR, a new publicly available dataset designed to advance smartphone-based human activity recognition (HAR) in office environments. PrevOccupAI-HAR comprises two sub-datasets: (1) a model development dataset collected under controlled conditions, featuring 20 subjects performing nine sub-activities associated to three main activity classes (sitting, standing, and walking), and (2) a real-world dataset captured in an unconstrained office setting captured from 13 subjects carrying out their daily office work for six hours continuously. Three machine learning models—namely, k-nearest neighbors (KNN), support vector machine (SVM), and Random Forest (RF)—were trained on the model development dataset to classify the three main classes independently of sub-activity variation. The KNN, SVM, and RF models achieved accuracies of 90.94%, 92.33%, and 93.02%, respectively, on the development dataset. When deployed on the real-world dataset, the models attained mean accuracies of 69.32%, 79.43%, and 77.81%, reflecting performance degradations between 21.62% and 12.90%. Analysis of sequential predictions revealed frequent short-duration misclassifications, predominantly between sitting and standing, resulting in unstable model outputs. The findings highlight key challenges in transitioning HAR models from controlled to real-world contexts and point to future research directions involving temporal deep learning architectures or post-processing methods to enhance prediction consistency.

This article presents PrevOccupAI-HAR, a new publicly available dataset designed to advance smartphone-based human activity recognition (HAR) in office environments. PrevOccupAI-HAR comprises two sub-datasets: (1) a model development dataset collected under controlled conditions, featuring 20 subjects performing nine sub-activities associated to three main activity classes (sitting, standing, and walking), and (2) a real-world dataset captured in an unconstrained office setting captured from 13 subjects carrying out their daily office work for six hours continuously. Three machine learning models, namely k-nearest neighbors (KNN), support vector machine (SVM), and random forest, were trained on the model development dataset to classify the three main classes independently of sub-activity variation. The models achieved accuracies of 90.94 %, 92.33 %, and 93.02 % for the KNN, SVM, and Random Forest, respectively, on the development dataset. When deployed on the real-world dataset, the models attained mean accuracies of 69.32 %, 79.43 %, and 77.81 %, reflecting performance degradations between 21.62 % and 12.90 %. Analysis of sequential predictions revealed frequent short-duration misclassifications, predominantly between sitting and standing, resulting in unstable model outputs. The findings highlight key challenges in transitioning HAR models from controlled to real-world contexts and point to future research directions involving temporal deep learning architectures or post-processing methods to enhance prediction consistency.