Understanding Fatigue Through Biosignals: A Comprehensive Dataset

Fatigue is a multifaceted construct, that represents an important part of human experience. The two main aspects of fatigue are the mental one and the physical one, that often intertwine, intensifying their collective impact on daily life and overall well-being. To soften this impact, understanding and quantifying fatigue is crucial. Physiological data play a pivotal role in the comprehension of fatigue, allowing a precious insight into the level and type of fatigue experienced. Though the analysis of these biosignals, researchers can determine whether the person is feeling mental fatigue, physical fatigue or a combination of both. This paper introduces MePhy, a comprehensive dataset containing various biosignals, gathered while inducing different fatigue conditions, in particular mental and physical fatigue. Among the biosignals closely associated with stress situations, we chose: eye activity, cardiac activity, electrodermal activity (EDA) and electromyography (EMG). Data were collected using different devices, including a camera, a chest strap and different sensors from the BITalino kit.


STUDY OVERVIEW 1.Introduction
Fatigue is a multifaceted condition that extends beyond the mere physical tiredness.Despite numerous attempts by researchers to describe fatigue, it still lacks a unique defnition, with many studies only capturing a small part of this broader and more complex concept [1].Fatigue entails a wide range of physiological and psychological factors that can afect an individual's ability to function efciently, leading to negative consequences on individual health and performance.
Due to its complexity, quantifying fatigue can be challenging.It can be measured through subjective measures, such as questionnaires, which are a simple way to gain insight into a person's emotional state.However, these measures are known to be subject to bias.On the other hand, physiological measures are unbiased and can be collected without interruptions, since the human body continuously produces physiological data [2].Physiological signals, such as electromyography (EMG), electrocardiography (ECG), electrodermal activity (EDA), electroencephalography (EEG), eye-blink rates, can be analyzed to identify the physical and mental state of a human being.
Two main types of fatigue have been identifed: mental fatigue and physical fatigue.The frst one is experienced during and after demanding cognitive activities and is characterized by feelings of tiredness and lack of concentration [3].The second one is induced by physical exercise, leading to a decrease in muscle power.
In this paper, we introduce MePhy, a dataset that includes physiological data collected while inducing mental and physical fatigue in the test participants.The signals were acquired in a non obtrusive way, using devices that did not limit subject's freedom to move.Hence, the approach used to collect data could be applied in real life scenarios, such an industrial working one.Given the numerous signals that can be obtained, we focused on four diferent ones: EMG, ECG, EDA, eye-blink rates, which are the most signifcant for fatigue detection [4][5][6].

Available Datasets
The goal of our experiment was to create a dataset that included various physiological signals collected while eliciting diferent types of fatigue.In previous works, mental and physical fatigue have been considered separately.The works presented in [9][10][11][12]  focused on eliciting mental fatigue.In every one of these, mental strain was obtained by demanding cognitive activities, such as mental arithmetic tasks and robot movement control.In [13] and [14] the focus is on the efects of extended physical fatigue on physiological signals: in the frst one, real industrial scenarios, like manual material handling and supply pick-up and insertion task, are simulated; the second one is based on aerobic exercises on a treadmill, a stationary bicycle and an elliptical trainer.In all of these studies, the only biosignal detected was the heart rate.Although mental and physical fatigue have been considered as independent, there is evidence that they interact.It has been proved that mental fatigue emphasizes the perception of physical fatigue and, at the same time, cognitive functions can deteriorate after enduring physical challenges [15].
The work presented in [16] focuses on detecting the occurrence of both mental and physical fatigue, based on the assumption that individuals are exposed to several sources of fatigue simultaneously.Test participants endured a four-part experiment that aimed to elicit mental fatigue, physical fatigue and a combination of both.
The only physiological signal detected was the heart rate, collected with Polar H10, a chest strap.
This dataset builds upon the considerations from [16].We included additional test subjects and new biosignals for analysis.Hence, the dataset provides a broader and more comprehensive view of biosignals and their correlation to stress and fatigue.
While reviewing diferent published works, we found a few datasets that best ft our study: WESAD [7], PASS [2], Fatigue-Set [3] and a dataset for stress recognition in automobile drivers [8].Among all the datasets considered, there are no signifcant diferences in terms of the collected biosignals, mainly ECG, EDA, EMG, but rather in the approach used to gather the data.
In particular, the WESAD dataset was focused on eliciting mental fatigue, not considering the physical component of fatigue.
The data available in the PASS dataset are achieved by inducing physical fatigue and afective fatigue, which is related to anxiety, discomfort or fear.The type of cognitive weariness that we induced during our experimental protocol is the mental fatigue, which refers to situations that require refection and problem-solving abilities.Hence, the data collected mirror two diferent conditions.
The main focus of the data collection used for the FatigueSet dataset was investigating the efect of physical activity on the generation of mental fatigue.Hence, they only considered the physical aspect in correlation to the mental one.
Lastly, the dataset presented in [8] concentrates on collecting physiological data during real-life driving tasks to defne a driver's respective stress level.In this work, they do not diferentiate between a physical component and a mental one, since they considered three general level of stress (low, medium and high).
A brief and intuitive comparison among these datasets can be found in Table 1.

METHODS
The aim of our experiment was to create a diverse and comprehensive dataset to study fatigue and its correlation with biosignals.To achieve this, we conducted an experiment to collect physiological data through diferent wearable devices, eliciting both cognitive and physical fatigue.The experimental protocol adopted is depicted in Fig. 1.
In this section, we explain the conducted experiment.

Experimental Protocol
The experiment consisted of four parts and in each one the participant faced a diferent condition: rest, mental fatigue, physical fatigue, a combination of both.The total duration of the experiment was approximately of 25 minutes.Before starting the experiment, participants were asked to wear the chest strap and the sensors were applied.The test was then explained to the subjects.The frst phase, which lasted 3 minutes, was used to record the baseline activity of the participant.The subject was kept at rest and seated, without exposure to any stimulus.
The second phase was dedicated to stimulate mental fatigue, using the same test as in [12].This test consisted in a combination of mathematics, memory and visual tasks that aimed to elicit cognitive efort and fatigue.An HTML-based application was developed and shown to the participants on a touch screen.We considered three types of stressors, presented for 1 minute each.They are shown in Fig. 2a.These tasks were presented in random order and some were also repeated.The frst task was the Stroop test [17] with fve diferent colours and it was repeated twice.The next task consisted in math calculation of multiplications with operands from 1 to 13.This test was presented twice, but the second time with operands from 9 to 20.Then, the 2-back test [18] was introduced to participants, consisting in a sequence of diferent images shown to the subject, who had to identify whether the current image had appeared two images back in the sequence.
The physical fatigue was induced during the third phase of the experiment.The participants were asked to perform isometric exercises for biceps and triceps contraction, which consisted in holding a dumbell with diferent angles of elbow fexion.The angles we used were 0°, 45°, 90° and 135° and are shown in Fig. 2b.Every position was held for a minute and they were repeated twice.Each participant chose a weight that challenged them (2kg, 3kg or 5kg), allowing them to complete the entire phase without stopping.The order of these two phases was inverted for half of the participants, in order to compensate for the efect of one type of fatigue on the other.
The last phase of the experiment aimed to induce mental and physical fatigue at the same time and, to do so, we combined the two tests mentioned before.Specifcally, test subjects were asked to perform the mental fatigue induction test while holding a dumbell in a sitting position, as in the physical fatigue induction test.We noticed that it was necessary for the subject to have both hands free to perform the tests correctly and at the same time, in order to carry out the physical fatigue induction test with one hand and to keep the other one still for data collection.For this reason, the participant had to answer the cognitive tests by voice, in order to have them entered by the experimenter.

Experimental Setup
The chest strap Polar H10 was used to record cardiac activity and it was connected through Bluetooth to a laptop, accessing data using the GATT (Generic Attribute Profle) protocol.Data were acquired with a sampling rate of 1 Hz.
Eye activity was recorded using Logitech C920 camera, in order to capture the frequency of opening and closing of the eyelids.The sampling rate used was 30 Hz.
As regards EDA and EMG, the BITalino kit was used [19].This kit includes: diferent sensors for biosignals acquisition, a microcontroller that converts analog signals in digital ones, and diferent types of cables.Data are acquired through a Bluetooth connection between BITalino sensors and a laptop and the software   OpenSignals is required to be able to interface with BITalino.The data were sampled at a rate of 1000 Hz.The literature identifes the palm region of the hand for extracting EDA and the trapezius muscle region for obtaining EMG.Electrodes were thus applied to these two body areas on the opposite side from the dominant hand (e.g.electrodes are placed on the left side if the subject is right-handed).
The placement of the sensors and the camera is shown in Fig. 3.

Test Participants
60 subjects (30 females, 30 males, mean age: 22,85) were involved in the experiment.The candidates were volunteer students, who were completely new to the experimental tasks and goals of the experiment.
For the frst half of the participants (30 subjects), we detected all of the biosignals presented in the paper; due to some issues encountered during signal acquisition, 11 recordings had to be removed, reducing the number of subjects with all four signals from 30 to 19.For the second half, the cardiac activity was the only biosignal collected.Nevertheless, the same experimental protocol was used in both cases.
Table 2 summarizes biosignals recorded for each subject.
The study protocol followed the Helsinki declaration and compliance to participate in the study was obtained from written informed consent before starting the experiment.All the data were reported anonymously.
ECG EDA EMG Eye Blinking user0 to user18 Table 2: Biosignals recorded for each subject

DATASET
During the experiment, we collected four diferent biosignals from the participants: heart rate, EDA, EMG and eye blinking.We then organized the retrieved data in a main folder, one for each subject.Inside the candidate folder, there are additional subfolders where we fnd the breakdown into each physiological signal collected for that subject.Finally, each subfolder contains four text fles in .txtformat, each corresponding to a condition elicited during the experiment.
The MePhy dataset can be found at 10.5281/zenodo.8423405.Data maintenance will be periodically carried out by the corresponding author.

Figure 1 :
Figure 1: Organization of the experiment.
(a) Mental fatigue inducing test.The test consisted in a combination of mathematics, memory and visual tasks executed following the sequence depicted in the fgure.(b) Physical fatigue inducing test.Each angle of elbow fexion was hold for 1 minute.

Figure 2 :
Figure 2: Organization of mental and physical fatigue inducing tests.

Figure 3 :
Figure 3: Placement of the sensors for EDA, EMG and eye blinking collection.The green circle highlights the position of the camera.The red circles indicate the region for obtaining EMG, while the blue circle the region for obtaining EDA.Polar H10 is not shown since it is worn under the shirt.

Table 1 :
Comparison among the available datasets considered.