The Effect of Demanding Mental Tasks on Electrodermal Activity and Heart Rate During Physical Activity: A Pilot Study

Received 20 April 2023, accepted 2 May 2023, date of publication 5 May 2023, date of current version 17 May 2023. Digital Object Identifier 10.1109/ACCESS.2023.3273303 The Effect of Demanding Mental Tasks on Electrodermal Activity and Heart Rate During Physical Activity: A Pilot Study JUAN A. CASTRO-GARCÍA 1,2, (Member, IEEE), ALBERTO J. MOLINA-CANTERO 1, MANUEL MERINO-MONGE 1, CLARA LEBRATO-VÁZQUEZ 1, AND ISABEL M. GÓMEZ-GÓNZALEZ 1,2, (Senior Member, IEEE) 1Departamento de Tecnología Electrónica, Universidad de Sevilla, 41012 Seville, Spain 2Instituto Universitario de Investigación en Ingeniería Informática, Universidad de Sevilla, 41012 Seville, Spain Corresponding author: Juan A. Castro-García ([email protected]) This work was supported by the Project AAI, funded by MCIN/AEI/10.13039/501100011033 of Spain’s Ministry of Science and Innovation under Grant PID2019-104323RB-C32. This work involved human subjects or animals in its research. Approval of all ethical and experimental procedures and protocols was granted by the Regional Government of Andalusia (Spain) under Protocol No. C.P. TAIS-C.I. 1130-N-17, 2018. ABSTRACT This paper presents the development of a virtual rehabilitation game and mental load with different difficulty blocks. The game was controlled with body tracking and physiological signals - electrocardiography and electrodermal activity - were recorded throughout the session. Several parameters - heart rate (HR), heart rate variability (HRV), skin conductance level (SCL), skin conductance response (SCR), energy expenditure ...- were extracted from these signals to check mental load influence on them. Mental load was found to affect the variation in kinetic power and instantaneous heart rate; a Support Vector Machine with linear kernel was trained with these two variables and an 82.3% accuracy rate was obtained. Furthermore, the mental load was reflected in the number of errors made by the volunteers, in the selection time and in the number of rounds in the game. INDEX TERMS Body tracker, ECG, EDA, mental load, rehabgame. I. INTRODUCTION According to the World Health Organization (WHO), when people do exercise frequently, there is a reduction in the risk of suffering from cardiopathies, diabetes, high blood pressure, breast and colon cancers [1]. WHO guidelines state that young people and adults need to do 150 min. a week of moderate to vigorous aerobic exercise [2]. However, some physical activity can be negatively affected by certain factors. One of these factors is stress. Some people, used to doing sports, use physical exercise to cope with stress, although, in the following study [3], its authors concluded that, stress actually has a negative effect, as it reduces efforts to perform Physical The associate editor coordinating the review of this manuscript and approving it for publication was Norbert Herencsar . Activity (PA). Stress elevates the concentration of cortisol in the blood and its negative effects on health [4] might be magnified when paired with a sedentary lifestyle, also implying an important risk factor related to obesity [5]. Stress can be defined as the perceived information overload as the feeling of being overwhelmed when environmental demands exceed the perceived capacities to cope with them [6], [7]. The most well-known cause is the work environment [8], but it also occurs during the performance of other daily activities such as driving a vehicle in urban environments [9]. To motivate and increase the frequency with which PA is carried out, virtual assistants can be used to guide the type of exercise and the number of repetitions to be performed [10]. These can be based on devices that monitor the subject, such as body trackers which have proven their efficiency VOLUME 11, 2023 This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ 46665 J. A. Castro-García et al.: Effect of Demanding Mental Tasks on EDA and HR During PA: A Pilot Study in many additional applications, such as video games [11], [12], interaction with computers [13], [14], cardiopulmonary resuscitation (CPR) [15], estimate of caloric expenditure [16], [17], [18], [19] or rehabilitation [20] in general. Additionally, the use of robots as tangible assistants to promote physical activity in the elderly [21] - also known as active aging - has also been tested, proving to be more motivating than virtual assistants [22]. While virtual assistants manage the rehabilitation session, other technologies, such as virtual reality [23] or the use of video games [24], can help with physical exercise. Numerous studies have shown that stress, as well as the performance of physical activities, are reflected in several physiological variables. Stress can generate variations similar to those associated with PA, such as blood pressure (BP), heart rate (HR) [25], [26], [27], [28], skin conductivity [29] and skin temperature [30]. This is why, the subject’s monitoring could be used with a dual objective: the first one being to ensure that s/he is performing the exercise; and the second one being to extract information about his/her general state and the level of stress in particular that the proposed activity is generating, and in order to adapt it to a level of effort better suited to the individual’s capabilities. This pilot study was carried out to determine the effect that stress or increased mental demand has on certain physiological variables during light PA -below 39% of the heart rate reserve [31]-, and to thereby identify an excess of mental load during its execution. A Kinect-based video game was therefore adapted in which different geometric shapes have to be reached on the screen by only moving both arms. The skeleton provided by Kinect is used to estimate the kinetic power (Pk) of the movements and consequently the energy expenditure. The electrocardiogram (ECG) and electrodermal activity (EDA) of both hands are recorded simultaneously. PA should be light, with intensity in the control phase being similar to the stressful phase, to avoid increases in heart rate due to energy demand and possible sweating due to the effect of thermoregulation. The further goal of this study is to verify if there are physiological variables that indicate the stress level of a user with physical disability during a gamified rehabilitation session; however, as they are a particularly sensitive population, this study was performed with a healthy population. II. STATE OF THE ART A state-of-the-art review focused on finding the answer to ‘‘stress during physical activity’’ was carried out following the guidelines described in [32]. The PRISMA flow graph, which summarizes the results obtained in different phases, is depicted in Figure 1. As an initial approach, we used the text ‘‘stress AND physical AND activity’’ in two databases: IEEE Xplore and Scopus, which returned 343 items in IEEE Xplore and 36859 in Scopus. Given the large number of results in the second database, the search was refined by adding the following keywords: ‘‘heart AND rate AND (gsr OR eda OR sc OR skc).’’ This more restrictive search returned 170 items in Scopus. FIGURE 1. Number of selected articles throughout the review process. We found an total of 13 duplicated papers that were removed, while the remaining 500 articles continued to the following two phases. In the first one, only the title and abstract were screened. This phase removed many studies about stress in physical materials. In the second phase, the remaining 112 articles went through a full-text reading process, in which papers that met one of the following excluding criteria were discarded: 1) The study does not include physiological signals. 2) It uses very few samples of a physiological signal (e.g., one single HR sample per day). 3) It uses animal models. 4) It is a review; and 5) It is a previous work from a selected paper. Finally, we selected 16 related papers whose main features are shown in Table 1. In the review carried out, we found that the most commonly used physiological variables for stress detection are derived from the recording of cardiac activity −87.5% of the cases-, in any of its variants: ECG [34], [38], [39], photoplethysmogram (PPG) [54] or blood volume pulse (BVP) [42], from which we obtain parameters based on the time or frequency domain, and in which Standard deviation of NN intervals (SDNN) is one of the most significant according to [41]. Secondly, we find EDA [42], [51], in 56.25% of cases, which has shown its efficacy when used alone or in conjunction with other variables, such as ECG [53], [57]. Less frequently, other authors have included physiological signals such as electroencephalography (EEG) [46], electromyography (EMG) for the measurement of muscle loading in the upper trapezius [46], BP [47], breathing rate (BR) [40], [41], maximal oxygen consumption (V02) [34] or skin temperature (ST) [41], [54]. To complement the physiological information, it may be advisable to include environmental variables, such as weather conditions [54], or the speed, position and state of the road, captured with a cell phone [42] while driving on a stretch of road; or those derived from the subject’s own PA, which can be measured with accelerometers [39], [53] or markers placed on the human body that are then captured by a camera to analyze the subject’s movement [52]. Accelerometers have also been used to limit the extent to which physiological information used to detect stress is corrupted by motion artifacts [33]. To complement the 46666 VOLUME 11, 2023 J. A. Castro-García et al.: Effect of Demanding Mental Tasks on EDA and HR During PA: A Pilot Study TABLE 1. Main features of the selected manuscripts obtained through the scoping review. VOLUME 11, 2023 46667 J. A. Castro-García et al.: Effect of Demanding Mental Tasks on EDA and HR During PA: A Pilot Study physiological information and to thus be able to classify it properly, validation tests of the state or level of stress to which the subject is subjected during the different phases of the experiment are needed. It is therefore known that, physiologically, a hormone, cortisol, is released in stressful situations. Some authors use periodic measurements of this hormone in saliva to check the stress level of the subject [33], [50]. However, the use of questionnaires to measure the emotional state is more widespread [34], [39], [42]. To induce stress many researchers have used a methodology in a controlled environment that seeks to exceed the subject’s normal processing load in different ways: by using arithmetic calculation tests [34], [40], [53], with time-varying restrictions; with an intelligence quotient [38]; seeking conflict in the identification of a word’s color that semantically expresses a color that is different from the color used to display it - Stroop test [53], [57]; with videogames [41], [51] where surprising elements are introduced that alter their course; by using virtual reality to simulate environments that induce extreme situations such as walking along a bar between two skyscrapers [52] or subjecting the subject to a stressful real environment such as speaking in public or defending an exam before a jury [42], [50]. There are also studies that seek experimentation in daily life environments [39], while driving [42] or with animal interaction [46]. The performance of PAs and how they affect physiological variables, in combination with stressful situations, is the main purpose of this study. In all the studies found, these experiments were carried out in controlled environments using a treadmill and involving walking at different speeds [50], running, uphill walking [34], among others ; balancing on a barbell [52], ergometer cycling [41], [49], [54] or just maintaining a posture - such as raising your arms 45◦and holding them up for several minutes [38]. Most of the studies analyzed seek to detect the existence - or non-existence - of stress, or different stress levels, through the use of classifiers. Among the most commonly used are k-nearest neighbors (kNN) [33], support vector machine (SVM), J48 [53], linear discriminant analysis (LDA) [46], artificial neural networks (ANN) [54], random forests (RF), decision tree (DT), or Bagging [39]. Several authors have studied the classifiers that generate the best results with the same dataset, while others combine them to improve prediction [42]. Feature space dimensionality reduction is a common preclassification task itself [42], [45] that can be used to determine which features are the most relevant. This is especially relevant for multi-modal systems that include several sensors that can extract multiple features. They do [41] in particular use the ridge logistic regression method to achieve this objective. The use of statistical techniques, such as the Wilcoxon test, allows us to determine the statistical significance between different stress states -or levelsaccording to selected variables or characteristics. In the literature reviewed, several authors have been found to use this analysis type [38], [57] or another analysis type aimed at finding correlations between the level of stress measured by questionnaires and these physiological variables [34], [49]. Classification accuracies range between 70% and 97.13%. Better results are generally obtained when EDA is added to the physiological variables. In [51], the authors obtained a sensitivity of 95%; in [50] an accuracy of 97.13% by combining EDA with ST; in [33] a correct stress classification was obtained in 93% of the participants; whereas a slightly lower accuracy of 92.4% was obtained in [53]. It was only in [57] that the accuracy obtained was between 70% and 85%. Stress generally produces an increase in skin potential response (SPR) [51] or skin conductivity [33], [49], [52]. Heart rate-derived information permits correlations to be established between stressful phases and pre- and post-stress phases [34], allowing us to differentiate between all combinations of exercise and induced stress [38]. Furthermore, if BR data is added, the accuracy of the system is improved [41]. Some studies affirm that adding other non-physiological inputs improves prediction [42], [53], [54], [58]. Not many studies have used EEG among the signals analyzed, but, in the existing ones, spectral characteristics such as high alpha power at the C4 position are highly correlated in more than 75% of the participants with the different stress phases [49]. Furthermore, the average EEG power and Mel Frequency Cepstral Coefficients (MFCC) were found to be the most significant features for classifying valence and arousal dimensions in human-horse interaction [49]. A decrease in evoked potentials has also been found during stressful exercise performance, suggesting an increase in cortical demand for primary task performance [52]. Combining PA and stress produces an increase in HR compared to cases when these situations occurred separately [38]. This means that it is possible to detect stress in a subject, regardless of whether they are resting or at work, with mental stress having a weaker effect than physical stress [47]. In [41], the authors obtained an accuracy of 74.5% in stress discrimination while exercising with an indoor cycle at different intensities, based on features derived from HR and BR. Moreover, significant differences between the PA phases and stress have been proven when other variables, such as EDA, ST, BP or Heat flux, are used [47]. III. METHODOLOGY This experiment is inspired in the idea of using games for rehabilitation (rehabgame). It is based on the use of Kinect v1 and a modification of ShapeGame, an application included in the Kinect package for Windows SDK v1.8. The participant’s skeleton is shown on screen along with different spinning geometric shapes. According to a cue shown on the screen, one of these shapes has to be reached by performing upper limb movements, with the lower limbs having to remain still. Some physiological variables were recorded during the experiment as we will explain below. Each participant attended a single session consisting of three activity blocks (Bn) plus a final rest period (R4) (Figure 2). Each block is also divided into three parts: 46668 VOLUME 11, 2023 J. A. Castro-García et al.: Effect of Demanding Mental Tasks on EDA and HR During PA: A Pilot Study FIGURE 2. Experimental timing. An initial rest period (Rn), a gentle activity level (RLn) and a stressful period (SLn). During rest periods, volunteers were asked to remain calm and it was suggested that they sat in a chair. If they sat down for 30 seconds before the start of the next part, they were instructed to stand up to prevent heart variability in the following part being influenced by the postural change [59]. Rest periods (Rn), as borders between consecutive activity parts, took 5 minutes - 300 sec. -, because EDA is a slow recovery physiological signal [60]. In the gentle activity level, (RLn) two shapes are shown on the screen: the cue, visible at the top; and the target, that can appear randomly in different positions, but always within the reach of the skeleton’s arms. Once a target is reached, a new one, with its cue, appears on the screen one second later. This delay allows the player to return to the initial position and perform similar movements for each target. The RLn part takes 90 seconds. In the stressful level (SLn) ten shapes appear distributed around the avatar and this takes 150 sec. Here, participants must search before selecting and be careful not to accidentally hit a non-cued shape. There are several geometric shapes that are identical to the cue, in a quantity that decreases as the experiment progresses. The total of PA per block times is 4 minutes, to avoid fatigue and keep it as light. FIGURE 3. Game pictures: RLiLevel on the left and SL1on the right. All blocks are basically similar, although with some subtle differences existing in two main points: 1) The maximum dwell time (DWT) in selecting a target; and 2) The number of available targets in the SLnpart. Both decrease as nincreases, guaranteeing that mental demand progressively increases. As reaching a target on screen takes 2 sec on average, this is the minimum DWT value configured for the ending block, and a new target is shown on the screen when the previous one is selected, a similar level of PA being expected FIGURE 4. Volunteer playing: left in rest period, center and right, during the activity. throughout the blocks, irrespective of the DWT established. If the participant does not select any shape before the DWT expires, or they select an incorrect one, an annoying sound will be played (negative feedback) and a new cue and target are then generated (a new round); there is a minimum number of rounds (Nmin R) even if the volunteer does not make any selections. The game also makes a warning sound if the player has correctly selected a target or approached a shape but not sufficiently to select it. This information is summarized in Table 2. Before the experiment, the volunteers played a demo in which, multiple valid targets identical to the cue, appeared at the same time. This training session lasted 1 minute. TABLE 2. Description of the main features of each level for all blocks: duration (D) in seconds, number of geometric shapes shown (Sh), number of identical shape targets (NT), and dwell time (DWT) in seconds, minimum number of rounds (Nmin R). The underlying idea of this experimental division is to establish comparisons between periods of similar physical activity (RLn-SLn) but with a different mental load. Differences between blocks stem from a progressive increase in mental load by means of reducing completion time and hindering the target search. We hypothesized that all RLnperiods have a similar level of PA and mental load, which are both higher than in the rest periods Rn. Significant differences must appear between SLnperiods with respect to their previous RLnperiods and between SL segments when increasing the mental demand. We also hypothesized that an increase in that cognitive load may slightly reduce PA, especially when the time constraints are more severe. Figure 5shows the expected effect of experimental blocks and periods on PA and cognitive load. A. DATA ACQUISITION Physical activity was obtained by recording the joint coordinates that Kinect v1 offers (Figure 6). Two physiological VOLUME 11, 2023 46669 J. A. Castro-García et al.: Effect of Demanding Mental Tasks on EDA and HR During PA: A Pilot Study FIGURE 5. Illustration of the expected effects on PA and cognitive demand throughout the experimental blocks. FIGURE 6. Kinect v1 joint list [66] and coordinate group used. signals were also acquired during the experiment: ECG and EDA. We used Ag/AgCl electrodes for ECG (Ambu White- Sensor 4200), placed according to Einthoven II lead positions and specific wet electrodes for EDA (Biopac EL507), opposed to dry ones [29], [61], which were placed on the thenar and hypothenar eminences of both hands (see Figure 7); however, we only used the EDA channel for the participant’s dominant hand. We also used PSM [62], an open-hardware shield placed on an Arduino Uno board, that was programmed with LBSP [63], an open-source library for real-time applications, in order to acquire data at a sampling frequency of 256 Hz per channel, and sending them wirelessly via Bluetooth. Data was acquired using LSLRec software based on the Lab-Streaming Layer (LSL) library [64], [65]. This software makes it possible to save the data together with different synchronization marks, thereby facilitating data identification and segmentation throughout the different phases of an experiment. B. PARTICIPANTS We initially gathered together a group of 29 healthy volunteers (16 men, 13 women), aged between 18 and 67 (mean 39.9 and standard deviation - SD - 14.2), and weighing between 50 and 120 kg (mean 74.7 and SD 16.4). However, FIGURE 7. Electrode placement diagram of the ECG (left) and EDA (right) electrodes. Two EDA channels were used, one in each hand. two men were excluded because some signals were not properly registered. This sample size is within the range of pilot studies [67], moreover, using Cochran’s equation with this population, a precision level of 15.3% with a confidence level of 90% was obtained. All participants agreed to take part in the experiment and signed the consent form. The Ethics committee at the Regional Government of Andalusia (Spain) approved the experiment (protocol code C.P. TAIS-C.I. 1130-N-17, 2018). C. PROCESSING This section describes the characteristics set analyzed. The sliding window technique was used for the time-series analysis at each block and level. The SLnlevels were divided into two segments of the same length as the RLnlevel in order to obtain the same number of samples. These two segments are placed at the beginning and at the end of the SLnperiod. In this way it is possible to analyze if any latency exists regarding the stressor effect onset, as well as avoiding possible biases. The first 90 seconds of the SLncan be considered the transient time, or the time of major changes in level, while the last 90 seconds can be understood to be when the physiological variables are relatively stabilized or in a steady state. Furthermore, in order to analyze the evolution in more detail, we study the variation in each of the following variables between the RLnlevel and its corresponding SLnlevel. To avoid the effect of outliers in every set of data associated to all the variables obtained, the method based on the interquartile range (IQR) was used, which considers outliers to be those values in the set that are out of the following range: [Q1−1.5IQR,Q3+1.5IQR], with IQR =Q3−Q1and Qi the i-th quartile. The Kruskal-Wallis test (KW Test) will be used to find statistical differences among sets. This is a nonparametric method that tests whether the samples originate from the same distribution and it does not assume a normal distribution of the residuals. 1) GAME STATISTICS The number of rounds (NR), number hit (H), missing targets (MT ), wrong targets (WT ) and selection time (Ts) are extracted from the game to compare the performance during the periods RLnand SLnin all difficulty blocks. 46670 VOLUME 11, 2023 J. A. Castro-García et al.: Effect of Demanding Mental Tasks on EDA and HR During PA: A Pilot Study TABLE 3. Percentage of body weight [19] and associated clusters for kinetic power estimate. 2) KINETIC POWER ESTIMATE Pk Power expenditure can be estimated by evaluating the kinetic power, Pk, of the arm movements - joints 5-12 in the skeleton provided by Kinect (see Figure 6). To do this, the mass and velocity of the joint’s coordinates are required. Velocity is easily obtained from the joint’s positions at two consecutive time points, while for the mass distribution in the body, we used the results published in [19], and summarized in Table 3. So, let s{.}be a set of joints that delimit an arm segment (forearm, and upper arm) or define the hand (see Table 3), and let rs{.}(t) be the position of its geometric center at time point t. The velocity, vs(t), of a body segment sat time tis given by Eq. 1, and let msbe the mass of segment s, obtained by applying the associated percentage to the whole body weight. Kinetic power, Pk, is therefore defined by Eq. 2, which includes the energy expenditure of all the arm segments. vs(t)=rs{.}(t)− rs{.}(t−1t) 1t(1) Pk=X s 1 2ms| vs(t)|2/1t s∈[s{8}s{12}s{6,7}s{5,6}s{9,10}s{10,11}] (2) The joint’s coordinates are low-pass filtered by means of an exponential smoothing filter that Kinect implements. Data were then segmented by a sliding window with a 50% overlap. 3) EDA FEATURES To process the EDA data, we used Ledalab tools [68], configured with Continuous Decomposition Analysis (CDA) to recover the characteristics of the underlying signal in the sudomotor nerve; with Standard trough-to-peak (TTP), which analyzes data window maximums and miminums; and with Global which offers general data values. Table 4- accessible via ledalab.de/documentation.htm - summarizes these variables. Recorded data were adapted to Ledalab input format by adding events at the time points when the first geometric shape of each set is shown. A two-second window (the smallest DWT) and a sensitivity of 1 µS were used. 4) ECG FEATURES ECG was used to extract the instantaneous heart rate (IHR) for each level of the game and the heart rate variability (HRV), TABLE 4. EDA-variables’ list. using 30-sec. windows with an overlap of 50% (ultra-short- term recording times [69]), which would allow us to obtain 5 windows at the shorter duration levels, (RLn), while for the IHR we used 6-sec. windows with the same overlap as in the previous case. The HRV has special interest given that it allows us to evaluate the activity of the autonomic nervous system (ANS) [70], and specifically the parasympathetic nervous system (PNS) and the sympathetic nervous system (SNS) whose activity can be affected in situations of high mental demand such as stress. From HRV we observed time characteristics such as Standard deviation of NN intervals (SDNN), the root mean square of successive differences between normal heartbeats (RMSSD) and the percentage of successive RR intervals that differ by more than 50 ms (pNN50); frequenty characteristics were not studied since the windows are very small [69]. IV. RESULTS A. GAME STATISTICS Table 5shows the averages of the four variables described in the previous section related to the activity performed by all participants. With respect to wrong targets (WTs), it can be observed during the SLnperiods that about 6.3% of failures occur in the first two levels while it almost doubles in SL3. There is a statistically significant difference between this level and previous levels (p<0.0011). Furthermore, WT cannot occur at RLnlevels, since a single target is displayed on the screen. The missing target (MT) obtained low values at all levels in the first blocks (<1%). Only in the last block did this error reach 10.78%, standing out from the other levels and blocks. This difference was statistically significant (p<0.0015). The MT number is non-zero at RLnlevels and somewhat higher for the last block, although without statistically significant differences. These results may be due to the Kinect skeleton not reaching the target properly some times, and requiring additional stretching that may not be done correctly in the pre-set time. In RL3, moreover, we have a DWT that is at the limit of how long it takes a subject to reach a target on average, so it seems reasonable that the MT number increased. As previously mentioned, there are no statistical differences VOLUME 11, 2023 46671 J. A. Castro-García et al.: Effect of Demanding Mental Tasks on EDA and HR During PA: A Pilot Study between the different levels and blocks, with the exception of SL3. As for the number of rounds (NR), Table 5shows the average for each level and block. Considering that the duration of the SLnlevel is longer than the RLnlevel, some normalization is required, dividing NRby the duration of the level. The normalized values obtained are specifically [0.46 0.40 0.49 0.42 0.51 0.46] following the same order of RL1-SL3, as shown in the table below. One may expect NRto decrease as the difficulty increases - the number of blocks advances - but the differences are not enough to make the regression slope statistically significant. However, note that a decrease in NR can be observed when moving from the RLnlevel to its SLn, these differences having a significant difference - p<0.001. Finally, in focusing on the selection time (Ts), an increase in time is noted between each SLnlevel and its respective RLn, making the difference between them increase as the block number advances, i.e. as cognitive load rises. All mean times show statistically significant differences between them, as well as showing a positive slope with increasing block number. TABLE 5. Wrong targets (WT) and missing targets (MT) in %, rounds, number of rounds (NR) and selection time (Ts) for all participants B. KINETIC POWER ESTIMATE As an example, Figure 8shows the behavior of variable Pk for participant S09, while Table 6shows the values obtained by all the volunteers. Note that the motion is poor at the Rn levels. For each block, the SLnlevel shows less motion with respect to its corresponding RLn, probably due to searching for the target among all the available ones. This difference was statistically negative in all blocks and subjects, 1Pk= PSLn k−PRLn k<0, yet, no significant correlation was found with the number of blocks, i.e., with increased cognitive load, so that no change in this difference can be confirmed with task difficulty. These results are in accordance with those obtained from NR. C. PHYSIOLOGICAL FEATURES As a first analysis, we compared the statistical significance of each proposed variable within each block Rnvs. RLnvs. SLn, and across the SLnsegments of the three blocks, in order to know if there is any effect on physical activity and/or cognitive demand in the variables. From the proposed characteristics, we selected those that were statistically significant for at least 90% of the total comparisons made between these levels and across all volunteers. TABLE 6. Mean and SD of PKfor all subjects and blocks. FIGURE 8. Subject S09’s Pkat each level using a 6 sec. data window. Table 7shows these percentages by level comparison using the Kruskal Wallis test (KW Test) with a significance value (p<0.05) for all EDA variables. The first and last 90 sec of the SLnsegment were also compared. As previously explained, the first 90 sec may reflect some transience, while greater stationarity is expected in the last 90 sec. Only the CDA.Tonic variable shows significant variations throughout the experiment, although another variable, Global.Mean, is close to the 90% limit. TABLE 7. Percentage of times each EDA variable obtained significant p-values at full levels, in the first 90 seconds (F90EDA) and the last 90 seconds (L90EDA) for all subjets. CDA.Tonic and Global.Mean, show an accumulation effect as the experiment progresses, as has been shown in Figure 9, and this can be observed more clearly in Figure 10 on the same participant. There is no return to the prior rest level when moving across the blocks, even though the rest period is 5 min. It would therefore be interesting to analyze how these variables change in the SLnsegment with respect to the previous one (1EDAn), and to determine whether these changes increase as the cognitive load grows. 46672 VOLUME 11, 2023 J. A. Castro-García et al.: Effect of Demanding Mental Tasks on EDA and HR During PA: A Pilot Study FIGURE 9. Subject S09’s CDA. Tonic and Global.Mean at each level with PA. FIGURE 10. IHR -black- and EDA -orange- for volunteer S09 during the experiment. Level transitions are marked in red. The 1CDA.Tonicnand 1Gobal.Meanndifference is positive for all subjects and blocks applying the bootstrap test [71] for a 5% significance level. These results were the same when we included the first 90 sec, the last 90 sec, or the whole segment of each level in the analysis. The linear regression shows that a positive slope exist for these variables with respect to the block number. This slope is statistically significant, which implies that as the difficulty increases, so does the 1s. This analysis of differences between segments and blocks was extended to the other EDA characteristics (Table 4). Only the 1TTP. Latency regression shows a significant negative tendency. This parameter measures the delay time between a new target appearing on the screen, and the detection of a pulse in the EDA, the so-called event-related skin conductance response (SCR). As difficulty increases, latency decreases. Regarding the features associated with HR, group differences have been observed between resting states Rnand any of the activity states - RLn,SLn. This was expected. Figures 10 and 11 show the S09 participant values for IHR. An increase in cardiac activity was observed during the PA segments with respect to those related to inactivity. Furthermore, unlike the EDA signal, the HR recovers in resting levels. Although Figure 11 shows a slight increase in HR values in the SLnsegments compared to their previous RLn, that difference was not significant for the majority of the participants and blocks. Such a significance was only observed in 1IHR between Block 1 and Block 3; and between Block 2 and Block 3 for all participants. No statistical differences were found in the HRV characteristics analyzed for the group. FIGURE 11. Subject S09’s HR at each level. V. DISCUSSION Indicators have been found to vary significantly during the different experimental phases. These include both physiological - 1Pk, CDA.Tonic, 1CDA.Tonic, 1Global.Mean and 1IHR - and non-physiological types - WT, missing target (MT), 1Rounds, Ts -; see Tables 5and 8. TABLE 8. Selected variable values between RLnand SLnfor the three experimental blocks and all subjets. In analyzing the game variables, all of them show significant differences between the SLnand its previous RLn. However, only selection time (Ts) shows a more game dynamic behavior, increasing in value as the difficulty of the task at SLnlevels grows (rising n). This was expected, as reducing the number of correct targets increases the search time. And, as the available time to perform this task in the SL3state was significantly limited, selection errors WT and omission errors MT grow, while the number of rounds achieved (or targets achieved) was lower compared to the previous levels. An analysis among all these variables allows us to observe the existence of a high and significant correlation (Pearson’s coefficient ρ=0.97, p=0.001) just for Tsand WT. VOLUME 11, 2023 46673