A new measurement for workload assessment in agricultural tasks: EDA-based real-time model

Abstract

Agriculture is vital for the global economy but remains one of the most hazardous industries due to the high prevalence of musculoskeletal disorders (MSDs). This study aimed to develop and validate an electrodermal activity (EDA)-based model for non-invasive, real-time assessment of upper limb task workload, addressing the impact of environmental factors on workers' mental and physical strain. To achieve this, participants engaged in simulated pruning and harvesting tasks, which are known for imposing substantial workload and injury risks on the upper extremities in agriculture. Anthropometric data, EDA signals, and the Borg rating of perceived exertion (RPE) were collected. A multinomial logistic regression (MLR) model was employed to classify workload levels based on EDA wavelet features and key task-related variables. The developed models for pruning and harvesting tasks explained 40–50 % of the variance in RPE, demonstrating the highest accuracy in the middle RPE group (88–89 %). Specificity was notably high (>91 %) across both tasks for low and high RPE groups. Additionally, the middle RPE group exhibited recall and F1 scores above 84.5 %, while the high RPE category demonstrated recall, precision, and F1 scores ranging from 73 % to 92 % for pruning and from 60 % to 75 % for harvesting. These findings underscore the model's potential for precise workload categorization and the development of effective management strategies. Furthermore, the proposed EDA-based framework may hold broader applicability across various occupational domains that require non-invasive and continuous workload monitoring.