Estimation of Work Productivity Using R–R Intervals and QRS Regions of Electrocardiograms during Computational Tasks under Cognitive Load

Abstract

In recent years, advances in information technology have markedly increased the proportion of intellectual work across various occupations. In this study, intellectual work is defined as tasks that involve receiving and judging external information, performing knowledge processing such as analysis and numerical computation, and generating outputs. Such work demands cognitive resources that are essential for understanding, retaining, and manipulating information. However, prolonged engagement depletes these resources, leading to reduced productivity and extended working hours. Taking breaks has been recognized as an effective countermeasure, and systems capable of recommending optimal break timing are anticipated. To realize such systems, objective visualization of work productivity is required, and prior studies have investigated biological signals for assessing cognitive load during intellectual tasks. For example, Yamaguchi reported that heart rate variability analysis of electrocardiograms (ECGs) during continuous addition tasks showed increased LF and LF/HF ratios—indices of sympathetic activity—and decreased HF, an index of parasympathetic activity. These findings suggest that ECGs are effective for evaluating cognitive load. Nevertheless, no established method currently exists for quantitatively estimating work productivity in intellectual tasks using biological signals.In this study, we selected computational tasks with cognitive load as representative intellectual work and developed an estimation model (the baseline model) that predicts productivity from ECGs recorded during task performance. This model achieved an R² of approximately 0.67 and an individual error rate of about 7%. Despite these results, its accuracy was limited, and the physiological basis of estimation remained unclear. To address these limitations, SHAP analysis was applied to identify ECG waveform components contributing to the model’s predictions. The analysis revealed that the model captured overall waveform morphology and that certain heartbeats contributed more strongly than others. Specifically, high-contribution beats were characterized by shorter RR intervals (RRI), larger QRS power, and higher R-wave amplitudes compared with low-contribution beats. These findings led to the hypothesis that RRI, QRS regions, and R-wave amplitude are key features for productivity estimation. Building on this insight, participant-specific models were constructed using convolutional neural networks (CNNs) to extract morphological and temporal features from ECG waveforms. Training results demonstrated enhanced performance, with an average R² of about 0.85 and a mean absolute percentage error (MAPE) of approximately 4%, surpassing the accuracy of the baseline model. These results indicate that RRI, QRS regions, and R-wave amplitude are effective indicators for estimating work productivity during intellectual tasks.

Keywords: work productivity, R-R Intervals, QRS Regions, Electrocardiograms, Cognitive Load

DOI: 10.54941/ahfe1006997