ISSN 2722-3213 articles 11, pp. 108 – 120, 2024 https://mechta.ub.ac.id/ DOI: 10.21776/MECHTA.2024.005.01.11 Copyright: © 2023 by the authors. 108 V05 N.1 CARDIAC BIOMETRICS AND PERCEIVED WORKLOAD REGRESSION ANALYSIS USING RANDOM FOREST REGRESSOR IN COGNITIVE MANUFACTURING TASKS Afifah Harmayanti 1) ✉, Ishardita Pambudi Tama 2) , Femiana Gapsari 1) , Zuardin Akbar 3) , Hans Juliano 4) 1) Mechanical Engineering Deptartment Universitas Brawijaya Malang, Indonesia afifahharmayanti94@gmail.com 2) Industrial Engineering Department Universitas Brawijaya Malang, Indonesia kangdith@ub.ac.id 3) Institute for Computational Design and Construction University of Stuttgart Stuttgart, Germany zuardin.akbar@icd.uni-stuttgart.de 4) Department of Electrical Engineering and Computer Science National Yang Ming Chiao Tung University Hsinchu, Taiwan hansjuliano.ee11@nycu.edu.tw Abstract Workload is crucial in managing and maintaining good performance of human resources and allocations. In an advanced manufacturing industry, human job functions had shifted to cognitive tasks. Thus, cognitive workload evaluation should be used to monitor worker’s workload in optimal condition. Most common tool of cognitive workload tools are perceived measurement, like NASA – TLX questionnaire. Despite of its sensitivity to capture workload felt by the workers, this subjective measurement was prone to bias. Objective measurement utilizing biometrics data of the human body during working state was useful to eliminate bias. Cardiac biometrics were one of the many that were closely related to mental activity changes. The objective of this study was to understand the relationship of cardiac biometrics to perceived workload as an indicator of cognitive workload analysis. The study utilized four biometrics, heart rate, HRV low frequency power, total frequency power and ratio of low and high frequency power, were used to analyzed a one hour long cognitive based study case. The study case was designed in a manufacturing planning context referring to manufacturing aptitude tests, to induce cognition process on 30 participants. The biometrics and NASA – TLX score result of all the participants, were then calculated as effect size standardization before input into random forest regressor model to analyze relationship between cardiac biometrics and perceived workload. The result found a moderate relationship between the two (r 2 = 0.576). Features importance also showed the most impactful feature to the model is the effect size of frequency power ratio. However, it is recommended to always consider evaluating multiple cardiac biometrics in workload analysis to ensure good model performance. Keywords: Cardiac Biometrics, Machine Learning, Cognitive Workload, Manufacturing, Feature Importance. 1. INTRODUCTION Evaluating workers’ workload is crucial in managing resources allocation and performance. In manufacturing industry, evaluating workload can be carried out in every part through all the entire supply chain process. Human factors affect the quality of whole production process [1] . Studies found relationship between human factors and human error [2,3] and occupational incidents [4,5] . Thus, monitoring human workload to ensure their working quality is important to maintain the quality of overall process.