The safety of construction workers is of prime importance in establishing an effective safety climate. Construction is unique to other industries due to its dynamic manifestation of challenges toward occupational health and safety on site. Unfortunately, the construction industry records the most significant number of fatalities when compared to other industry sectors in the United States (BLS, 2021). Among others, falls, caught-in/between, struck-by, and electrocution, famously known as fatal four , are identified as primary causes of fatalities in the construction sector, with associated fatalities surpassing 6,000 between 2016 and 2020 (CPWR, 2022). Several studies have attributed These unfavorable statistics to industries' complex, dynamic, and demanding nature (Albert et al., 2017;Chi et al., 2005;Namian et al., 2016;Uddin et al., 2020). The dynamic nature of the construction workplace introduces a wide range of safety hazards that must be identified and mitigated for comprehensive workplace safety. The Construction Fatal Four program, an initiative by the Occupational Safety and Health Administration (OSHA), aims to improve hazard recognition and reduce injury rates by addressing common causes of fatal incidents affiliated with fatal four (OSHA, 2011). Furthermore, near-miss identification and reporting are preventive measures that can extensively mitigate workplace accidents and fatalities. OSHA defines a near-miss as an incident that does not result in injury or fatality but could have occurred given a slight shift in time or position (OSHA, 2002). Physical fatigue induced by physical exertion is a mundane part of construction, as 90% of construction tasks involve manual handling (CPWR, 2018). Among construction workers, fatigue induced due to over-physical exertion is often characterized by extended work hours/shifts without rest, a confined workplace without proper ventilation, difficult weather and working situations, and physically challenging/demanding activities (Anwer, Li, Antwi-Afari, Umer, & Wong, 2021;Aryal et al., 2017;Umer et al., 2020). Physical fatigue is listed in the top ten critical risks experienced by construction workers and has a high probability of causing accidents (Chan, 2011;Longo et al., 2019). This risk exposure increased the error rate among workers' regarding their tendency for unsafe behavior on site (Fang et al., 2015).
Therefore, this study uses physiological and personality parameters to investigate how physical exertion affects workers' near-miss recognition performance. Wearable sensors were used to capture and evaluate participants' physiological parameters of heart rate, heart rate variability, and electrodermal activity during the experimental trial to measure the presence of physical fatigue due to overexertion and variability during recognition activity. Eye-tracking technology and near-miss stimuli images were used to objectively test participants' near-miss recognition performance pre and post-physical exertion. The efforts of this study will provide theoretical and practical implications regarding the impact of physical exertion on workers' safety behavior regarding near-miss detection. It will also supplement the development of personality safety assessment and exertion relief protocols for highly susceptible workers to risk-taking behavior.
According to Heinrich's near-miss pyramid, the ratio of accidents to near-misses is 1:10, whereas according to Bird's research, this ratio is 1:15 (Frank, 1974;Heinrich, 1931). Near-miss events are categorized as precursors to an accident that could have resulted in injury or loss of life but did not. Identification of these near-misses facilitates organizations in avoiding future hazards that can result in non-fatal or fatal accidents. Therefore, near-miss recognition is increasingly vital in safety management systems to protect construction workers (Dillon et al., 2016). Due to the critical role near-miss recognition plays in predicting recordable incidents or fatalities, improving workers' recognition capabilities is vital for a thriving safety climate. Past research shows that implementing safety-specific transformational leadership can enhance workers' nearmiss recognition by enhancing learning goals (Lu et al., 2019). Similarly, near-miss recognition among construction workers can be improved if their identification performance is quantified and evaluated using state-of-the-art technologies such as visual sensing-driven eye-tracking technology.
Eye-tracking technology has been utilized in various fields of study to facilitate research through visual sensing, including transportation safety, clinical psychology, neuroscience, behavioral research, and construction safety. A study was conducted to evaluate the influence of visual cues on workers' hazard recognition behavior, cognitive load, and adaptation level during safety training adopted eye-tracking in their experimental data collection (Fu et al., 2024). In the field of construction safety, the use of eye-tracking technology to explore the impact of time pressure and mental demand on workers' attention distribution and cognitive process during high-risk electrical activity articulates the usability of fixation-derived matrices as vital indicators of selective attention (Pooladvand & Hasanzadeh, 2023). Additionally, research was conducted to examine the relationship between mental workload and environmental changes and their impact on an individual's behavior using driver-eye movement (Foy & Chapman, 2018). In clinical nursing, visual sensing through eye-tracking has been utilized to evaluate nurses pattern-matching failures and inattentional blindness with their fixation and dwell time (Al-Moteri et al., 2018).
Essentially, limitations in safety behavior result in human error that can reduce workers' attention to recognizing potential risky incidents on the job site that could lead to accidents and injuries (Hasanzadeh et al., 2017). Fixation duration is one of the most used parameters that evaluates an individual's time on a target area of interest after a visual search is initiated (Comu et al., 2021). Conceptually, a more significant fixation duration corresponds to better attention allocation. Similarly, the time to first fixation is when individuals' first fixation landed on the target area of interest during visual search activity; this parameter provides insight into individuals' adaptation and prioritization ability during the recognition process (Kang et al., 2024;Zhang et al., 2021).
Construction environments are notorious for workplace stress due to their intertwined and dynamic nature (Leung et al., 2012). Statistically, working in demanding and stressful environments results in nearly 70% of construction professionals suffering from excessive stress (Liang et al., 2022). Constant exposure to stressful conditions can alter workers' safety behavior, increasing risk-taking activities. Workers' safety behavior is vital in establishing an optimal safety environment on construction sites. The site's physiological, psychological, or environmental conditions can alter this safety behavior. When workers are exposed to external stress, their risk propensity and perception vary from their natural state. Some everyday stresses studied include mental, auditory, environmental, psychophysiological, and physical stress (Min et al., 2012).
The most impactful alternator in workers' safety behavior is physical stress induced by extended physical exertion-induced overexertion. A study conducted to evaluate safety behavior and safety management suggests that the disproportionate rate of construction injuries and accidents might be due to human-related factors such as excess physical exertion and physical fatigue (Langford et al., 2000). Additionally, a study revealed that most work-related injuries in the construction sector are due to overexertion-led fatigue (Chan, 2011). Physical fatigue induced by physical exertion is usually measured by subjective survey and physiological metrics, but the dynamic nature of construction job sites makes it challenging to measure fatigue on the job sites in real-time (Anwer, Li, Antwi-Afari, Umer, & Wong, 2021). Hence, wearable sensors used to collect bio-signals have been widely adopted in non-clinical trials to objectively track physiological changes induced by physical fatigue in real time (Anwer, Li, Antwi-Afari, Umer, & Wong, 2021;Jebelli et al., 2019).
Among the physiological parameters, heart rate (HR) is the most prolific replication of a human body's response to exertion induced by external stimuli in the form of physical or mental fatigue (Kumar et al., 2007). In several studies, researchers have measured the heart rate of construction workers to analyze the relationship between physical stress and heart rate, providing evidence of increment in heart rate measurement when there is an increase in physical stress induced by high demands of work tasks of environmental conditions (Gatti et al., 2014;Ueno et al., 2018). Electrodermal Activity (EDA) has proven to be a reliable physiological indicator that monitors workers' internal reactions using skin moisture level fluctuations when the human body is exposed to specific events that produce different levels of electrical signals. Several studies have explored the use of EDA data as a surrogate measure of workers' fatigue levels during construction activity designed to induce physical fatigue in controlled and actual on-site environments (Anwer, Li, Antwi-Afari, Umer, Mehmood, et al., 2021;Anwer et al., 2020). Using EDA data collected through wearable sensors to determine workers' physical demand by monitoring their physical fatigue provides non-invasive validated techniques to improve fatigue assessment in construction (Aryal et al., 2017;Lee et al., 2021;Umer et al., 2020). In addition, heart rate variability (HRV) is also considered an accurate indicator of physiological changes controlled by the autonomic nervous system (ANS). The use of HRV to objectively measure physical exertion-induced fatigue was achieved in a study where 80 healthy subjects' HRV data was collected during preset treadmill exercise. 11 out of 24 features extracted from HRV data were used to classify and model physical fatigue classification using machine learning algorithm (Ni et al., 2022). Similarly, another study evaluated the usability of the HRV parameter for physical exertion modeling during manual material handling construction task-induced fatigue (Umer et al., 2020).
According to the 40-item mini-marker, the Big Five personality model developed by Saucier based on Goldberg's unipolar Big-Five markers describes personality by five factors: Extraversion (tendency to experience positive emotion and gregarious), Agreeableness (propensity to be cooperative and empathetic), Conscientiousness (practitioner of systematisms and organized), openness/intellect (curious tendency and imaginative nature), and neuroticism (temperamental nature and moody) (Gerard Saucier, 1994).
As discussed in the previous section, physical exertion shows correlation with workers' safety behavior. Therefore, it is vital to investigate personality traits to understand workers' safety behavior toward unsafe acts to provide preventive measures. In recent years, there has been an increased effort to understand how different personalities affect workers' behavior toward hazard identification (Hasanzadeh et al., 2019;Pourmazaherian et al., 2021). In addition, the impact of personality on near-miss recognition performance has also been evaluated providing correlation of construction workers personality traits and their close call detection capabilities (Muley et al., 2024). Although, Big Five personality traits such as Agreeableness, Conscientiousness, and Openness/Intellect have association with workers safety awareness (Ma et al., 2021;Muley et al., 2024;Pourmazaherian et al., 2021) neuroticism and extraversion have been identified in previous studies as an indicator for predicting physical health symptoms for an individual (Klinger-König et al., 2018). Furthermore, high neuroticism is related to lower aerobic capacity and lower energy expenditure during a challenging physical activity. This lower energy level within individuals with high neuroticism personality traits contributes to higher physical exertion and increased overall fatigue compared to other individuals (Terracciano et al., 2013). Additionally, high extraversion personality individuals are associated with better physical and mental outcomes like lower frailty, lower sleep difficulties, and fewer depressive symptoms that manifest into lower exertion and fatigue. This trend is also observed in individuals with high conscientiousness, openness/intellect, and agreeableness (Hakulinen et al., 2015;Stephan et al., 2018;Stephan et al., 2017). Personality traits are also associated with an individual's behaviors that have implications for fatigue. A study evaluating the relationship between personality traits and a sedentary lifestyle using a meta-analysis of 16 samples (N>125,00) showcases that higher neuroticism is related to physical inactivity and sedentary behaviors. In contrast, higher extraversion, openness, agreeableness, and conscientiousness are associated with a physically active lifestyle (Sutin et al., 2016). Hence, based on results from previous studies, extraversion and neuroticism personality traits were further explored to understand better workers' safety behavior for near-miss identification relationship with individuals' physical attributes of physical exertion.
In construction safety, the biggest challenge is maintaining optimum work efficiency without any lags in safety. Construction, a physically demanding industry, exposes workers to high-intensity physical exertion daily on-site. Managing and reducing physical exertion is vital to keeping workers safe and eliminating any negative impact that can lead to unsafe behavior. Research conducted to evaluate the influence of physical exertion on workers shows that fatigue reduces workers' cognitive ability and decision-making skill set and lowers their ability to process hazards, resulting in unsafe behavior and an increased error rate (Ibrahim et al., 2023;Namian et al., 2021;Qu & Guo, 2019;Taherpour et al., 2021). Although these studies have explored the impact of fatigue induction on workers' risk-taking behavior, safety attitude, and situational awareness, their central focus has been on using construction hazards and workers' subjective rating as points of interest, ignoring the precursor of any near-miss incidents. Therefore, investigating the impact of physical exertion on the near-miss recognition performance of participants can provide vital information regarding workers' cognitive ability and safety awareness when exposed to physically exerting activity (figure 1).
• H1: Exposure to physical exertion reduces workers' near-miss recognition performance. The use of wearable biosensors to track workers' overexertion induced by physical exertion using physiological measurements like heart rate (HR), heart rate variability (HRV), and electrodermal activity (EDA) in real-time has showcased the potential to improve safety on job sites (Anwer, Li, Antwi-Afari, Umer, & Wong, 2021;Aryal et al., 2017;Umer et al., 2020). In the case of HRV, the time-domain measure of the root mean square of successive RR interval differences (RMSSD) exhibits promising measurement of physical fatigue (Schmitt et al., 2015). The phasic component of EDA (EDR) is considered a reliable evaluation parameter that reflects rapid automatic changes in EDA due to external stressors such as physical exertion (Boucsein, 2012). Although research has been conducted to explore the association between physiological parameters and hazard recognition ability, there is still a limitation in linking these parameters with near-miss recognition ability, which is the precursor to any hazard on site. Therefore, this study used three physiological indicators, HR, RMSSD, and EDR, to examine participants' physiological changes due to physical exertion during near-miss recognition activity. The association of workers' bio-signals with their near-miss recognition efficiency under exertion can provide inference on the usability of physiological parameters to predict workers' safety behavior and preventive attributes.
• H2a: Exposure to physical exertion does not affect participants' EDR.
• H2b: Exposure to physical exertion affects and increases participants' HR.
• H2c: Exposure to physical exertion does not affect participants' RMSSD.
The safety behavior of construction workers is influenced by the personality characteristics they showcase (Xia et al., 2021). Therefore, the interdependency of personality traits and safety behavior in construction has been explored in previous studies that provide evidence of personality traits being highly aligned with the safety attributes of construction workers. For example, highly extroverted workers are associated with risk-taking behavior and reduced safety compliance (Barrick et al., 2013;Golimbet et al., 2007).
Neuroticism personality traits are also negatively affiliated with safety performance; highly neurotic individuals demonstrate lower safety compliance and participation (Conchie, 2013;Neal & Griffin, 2006).
Additionally, the influence of extraversion and neurotic personality traits on physical fatigue and activity determination showcases that individuals with low extraversion and high neurotic personality traits are more prone to be affected by overexertion induced by physical exertion activity. Therefore, exploring the influence of these personality traits on visual attentiveness and adaptation levels for construction workers' near-miss recognition performance with overexertion exposure can provide valuable insights for fatigue management protocols for workers with a high risk of failure in safety compliance.
• H3a: Exposure to physical exertion does not influence near-miss recognition performance of low or high-extraversion personality participants • H3b: Exposure to physical exertion does not influence near-miss recognition performance of low or high neurotic personality participants
Physical exertion can negatively impact construction workers' cognitive ability, leading to unsafe behavior on the job site (Zhang et al., 2023). This increase in unsafe behavior can increase perceived risk and reduce workers' near-miss recognition ability. The research methodology for evaluating the influence and impact of physical exertion on near-miss recognition performance is showcased in (Figure 2) In addition; this research methodology implements physiological parameters and personality traits to evaluate the impact of physical exertion on near-miss recognition performance, providing valuable insight into workers' safety behavior under physical stress.
The Borg-20 is a widely used subjective measure to evaluate the rating of perceived exertion during and after physical activity is performed. It is a 6-to-20-point scale, where 6 corresponds to "No physical exertion at all," and 20 indicates "Maximal exertion." (Borg, 1982).
Participants personality traits were collected using 40-item mini-marker personality test (G. Saucier, 1994). This personality test qualifies for its robustness and time constraints. All participants can complete this test in approximately 5 min with the ability to produce reasonable Big Five factors in small samples (Wood et al., 2010). This subjective evaluation of individual personality includes 40 adjectives used to describe yourself on a 9-point Likert scale.
In this study, the E4 Empathic Wristwatch, an off-the-shelf wristband type sensor manufactured by Empatica Inc, Cambridge, MA, US (Figure 1a), was used to collect EDA data of participants while performing experimental trials. EDA electrodes under the snaps were on the non-dominant wrist's inside and aligned with the participant's third (ring) finger. This wearable sensor records EDA data at a sampling rate of 4 Hz.
Polar H10 heart rate sensor manufactured by Polar Electro, Inc., Finland is a wearable heart rate device with a data collecting frequency of 1Hz and operating temperature ranging from -10 ○ C to 50 ○ C. The polar H10 heart rate sensor has been considered a gold standard of high precision and accuracy for heart rate monitoring (Chattopadhyay & Das, 2021). This wearable sensor is positioned at the chest level right at the end of the sternum bone, making sure the sensors touch the skin of the chest to record the heart rate in realtime.
To collect the eye tracking matrix during the near-miss recognition experiment, Tobii Pro Glasses 2 were used. Wearable eye tracking technology is valid in quantifying visual sensing and attention allocation data to gather and analyze the two-dimensional eye-movement-related information of participants on the areas of interest (AOIs). Tobii Pro Glasses 2 (manufactured by Tobii AB Company, Danderyd Municipality, Sweden) provides a gaze sampling frequency of 100 Hz and camera angle ranges of 82° (horizontal) and 52° (vertical) with 1-point calibration ability.
Figure 2: Experimental Methodology
The Louisiana State University's Institutional Review Board (IRB) approved this study, in which 42 healthy adult participants were recruited. This recruitment was based on the inclusion and exclusion criteria shown in table 1. All participants were required to complete a demographic, construction experience, and medical history questionnaire. The pre-screening process excluded seven participants who did not satisfy the inclusion criteria and have past medical history and incomplete fixation data. The remaining thirty-five participants (5 females) completed the study. Most participants recruited in this study were construction management students, with 23 participants having 1-5 years of industry experience and 2 participants having more than five years of experience. In addition, 8 participants reported exposure to past injuries out of 35. Power analysis was conducted to estimate the sample size for the study design using G*power 3.1.9.4 (Faul et al., 2007). With a conservative effect size of f = 0.25, alpha at 0.05, and power at 0.80, the projected total sample size needed with respective effect size is approx. N = 32. Therefore, this study's sample size (N=35) satisfied power analysis estimation. In addition, past research conducted on safety behavior, nearmisses, and eye-tracking experiments (Hasanzadeh et al., 2019;Ibrahim et al., 2023;Muley et al., 2024), 28-30 participants would have provided satisfactory results. Therefore, in this study we included additional 5 participants to reduce any potential attrition.
In this study, a repeated measure experimental design was adopted to evaluate the impact of physical exertion on the near-miss recognition performance of construction workers. The experimental research comprised five steps: The preparation stage included participant recruitment and collection of nearmiss stimuli scenarios. Conducting the experimental procedure involving non-stress and physical exertion trials, each of which was followed by eye-tracking activity, data processing and evaluation, statistical analysis, and experimental implications. The non-stress trial comprised participants viewing a documentary titled "World Class Trains -The Venice Simplon Orient Express" while seated in a relaxed position. This documentary is recognized as emotionally neutral, and researchers have confirmed that watching it does not induce any external stress on the viewer (O' Keeffe et al., 2020;Umer et al., 2022). In comparison, the physical exertion trial included participants completing a manual material handling (MMH) task comprising of lifting the box of 15 lbs. with box dimensions (45.72 cm x 30.48 cm x 17.78 cm) from floor level in a squatting posture, carrying it through a designated walkway path of 5 m by 3.66 m, and placing it on a shelf at shoulder height lifting (figure 3) and (figure 4) to induce physical exertion. This task was selected based on past study that evaluate construction workers physical exertion using physiological parameters (Umer et al., 2020). The box had ergonomic handles to avoid tangential discomfort in hands and fingers, whereas the MMH weight of 15 lbs. was selected based on the maximum acceptable weight assessed using interpolation from the Snooks table for males and females (Snook & Ciriello, 1991). Additionally, the NIOSH lifting equation was used to evaluate the risk of the MMH task. Upon analysis, the Recommended Weight Limit (RWL) provided a Lifting Index (LI) of 0.85, indicating that the MMH task was within a safe limit to be carried out by male and female participants (Waters et al., 1993). All trials were conducted in an ergonomic-friendly laboratory with controlled environment parameters. Figure 4: Physical Activity Environment and manual material lifting (A-Lifting Setup, B-Box with 15 lbs. weight, C-Carrying Trajectory)
Stimuli images used during eye-tacking tasks to evaluate the near-miss recognition performance of participants were collected through web-scrapping, case images from training, and pictures taken from real construction sites. To evaluate near-miss identification ability, 28 images showcased fatal four near-misses; additionally, eight images with no near-miss were included as false positives. These sets of selected images were manually processed to mark areas of interest (AOIs), which EHS and construction experts examined for rectification and vetting of assigned AOIs (figure 5). The near-miss event is categorized as an incident that may or may not lead to an accident or loss of property. Therefore, a near-miss scenario is usually considered a precursor to an incident that simultaneously showcases a potential injury-prone person and an injury-causing hazard at any instance. Special care was taken into consideration to collect images that show potentially fatal four hazards within the vicinity of construction workers to showcase near-miss events in static form.
Age 18-50 years Sex
Men and women (non-pregnant by self-report) PAR-Q All "no" answers on the questionnaire
The capability and willingness to give written informed consent and to understand the exclusion criteria.
Eye Problem Wearing eyeglasses or contact lenses Eye surgery: corneal (e.g., LASIK, RK), cataract, intraocular implants Eye movement or alignment abnormalities: lazy eye, strabismus, nystagmus Mental Problems Mental illness within the past five years or currently undergoing treatment for any mental illness. Exercise-Induced Asthma Exercise-induced bronchoconstriction is when the airways narrow and develop asthma symptoms because of exercise. Preexisting Spine Injury, Vertebral Deformities, or Musculoskeletal Disorders Slip-Disk, Anterior or Central Cord Syndrome, Pinch nerve, Muscle Spasm are some of vertebral or Spine illness, a low-back injury, or recent pain will be excluded from the experiment. Blood pressure Known systolic blood pressure >160mmHg and diastolic blood pressure >100 mmHg or taking medications that alter heart rate responses to exercise. Medications Blood pressure medications, beta-blockers, or any other medications are associated with increased risk of cardiovascular disease or indicative of metabolic, pulmonary, mental, or cardiovascular disease.
Including but not limited to chronic or recurrent respiratory, gastrointestinal, neuromuscular, neurological, or psychiatric conditions. Hospitalization for Spine illness within the past one year or currently undergoing treatment for severe Vertebral illness. Musculoskeletal problems were interfering with the exercise. Autoimmune or collagen vascular diseases. Any other medical condition or disease that is lifethreatening or that can interfere or be aggravated by exercise.
Table 1: Inclusion and Exclusion Criteria for the Study
This study addressed four individual biases to eliminate experimental design biases that could have altered analysis and results.
• Carryover effect: The carryover effect occurs in experimental design when participants perform multiple segments of the experiment continuously, which can enforce altering impact on results.
To avoid this, both trials were conducted on two separate days with a minimum duration between the non-stress and physical exertion trial of 24 hrs. • Skill level effect: Participants' skill level in the form of variable safety knowledge to identify four near-miss fatalities is a confounding factor that can impact the outcome of this study. To control this bias, participants were requested to view near-miss and fatal four training videos to ensure each participant understood how and what to identify during the eye-tracking task. The desired passing rate for the complete training evaluation quiz was 70%. • Selection Bias: Selection bias occurs when selected stimuli images do not have any characteristic differences when assessed by participants over time. Randomization of selected near-miss stimuli images was adopted to remove selection bias for study design. Two different sets of stimuli images were created, each with 18 images. Therefore, the experiment exposed the participant to specific stimuli images only once. • Order effect: To remove order effect bias from the study design, each participant completed nonstress and physical exertion tasks in a randomized order that was allocated.
In this study, the experimental procedure (figure 6) began with a post-recruitment task. All participants were required to complete a demographic questionnaire, a Big Five personality survey, and safety training before arriving at the laboratory. Once they arrived, participants were briefed on the experiment protocol, and their consent was obtained to participate electronically via a consent form. Participants then underwent a training evaluation quiz; those who scored below 70% were asked to review the safety training video until they achieved the desired passing rate.
Next, sensor setup and calibration were performed. Each sensor was placed at its designated location on the participants and connected to respective wireless devices to initiate physiological and visual sensing data collection. After calibration, participants were instructed to sit in a relaxed position before the commencement of the first trial. During this time, baseline heart rate (HR) and electrodermal activity (EDA) data were collected for 5 minutes to observe variations visually.
Assuming participants began the experiment with a non-stressful trial on their first visit, they were instructed to watch an emotionally natural documentary while seated in a relaxed position for 10 minutes, during which their HR and EDA data were collected. The control activity concluded at the end of the 10minute mark, after which participants engaged in a "fatal four" near-miss recognition eye-tracking activity. Participants were positioned 45cm in front of a monitor and asked to observe each stimulus image on the screen for 15 seconds. Post-observation subjective confirmation of near-miss presence was recorded. The task duration varied for each participant, but it took 7-10 minutes on average.
During the second visit, the trial began with sensor setup and calibration. Participants then performed a 10-minute session of physical exertion activity. Physiological data (HR and EDA) were collected throughout this manual material handling activity. Additionally, the Borg Scale was utilized to gather subjective assessments of physical exertion at the beginning of the activity and at every 2-minute interval. After completing the physical exertion activity, participants repeated the near-miss recognition activity. Once all tasks were completed, sensors were removed, and participants were free to leave. The entire experiment was conducted over two separate days, with the first visit lasting 30 minutes and the second lasting 25 minutes.
To evaluate the near-miss recognition performance of participants, eye-tracking matrices were extracted for each target AOI from stimuli images using Tobii Pro lab software. To verify the utility of this objective measurement for near-miss recognition performance based on visual sensing, a subjective assessment using near-miss recognition performance (NMRP) was also conducted. During eye-tracking activity for nonstress and physical exertion trials, participants completed a questionnaire regarding the presence and type of near-miss after each stimuli image. Each eye-tacking activity included 14 stimuli images with 16 wellbalanced fatal four near-misses (4-slip/Trip/Fall, 4-Struck By, 4-Caught-in-Between, 4-Electrocution). The minimum number of fatal four near-miss on each image was 1, while the maximum was 2. The NMRP of each participant for each stimuli image was calculated using equations 1 and 2. Subjective near-miss recognition assessment also evaluated the inclusion and exclusion of fixation matrices based on participants' awareness of identified near-miss. Additionally, to verify the usability of fixation matrices as indicators of recognition performance, correlation analysis was conducted between NMRP and fixation-derived matrices of fixation duration (FD) and time to first fixation (TFF).
Physical exertion detection and measurement are vital to predict if an individual is experiencing overexertion. To detect the presence of physical exertion in participants during and after the physical exertion trial, this study utilized Borg's Scale as a subjective assessment tool and physiological parameters HR and HRV, which were collected using Polar H10 as an objective assessment tool. HR and HRV have been used as prominent parameters to track cardiovascular activity and exertion (Schmitt et al., 2013). HR is identified as the number of heartbeats per minute, whereas HRV is the difference in time between two consecutive heartbeats. Some researchers suggest that healthy individuals will have high resting HRV, which increases their ability to adapt to unanticipated situations, whereas individuals with low HRV tend to decrease their physical and cognitive performance (Meeusen et al., 2013;Torres et al., 2008). Exposure to physical exertion is evaluated using age-predicted maximal heart rate (APMHR) expressed in equation 3. Additionally, to assess the intensity of physical activity per the American College of Sports Medicine (ACSM), low-intensity physical activity should be between 50 and 63% of HR max , between 64 and 76% for moderate intensity, and vigorous intensity of physical activity should be between 77 and 93% of HR max (Chen et al., 2020). The maximum HR for each participant during physical exertion activity was extracted during the HRV decomposition process using the hrv-analysis Python package (Champseix, 2021).
Based on the HR max of participants during physical exertion activity, 85% experienced moderate to vigorous physical activity. In addition to the objective exertion rating, this study utilized Borg's scale to capture perceived exertion (RPE) ratings during physical exertion tasks. Correlation analysis between RPE, physical exertion intensity, and HR were conducted to verify the utility of RPE to detect physical exertion.
Heart rate data for this study was collected using a Polar H10 heart rate sensor. Polar H10 is considered a reliable device by researchers as it is specifically designed to measure the cardiac output activity of the human body. Heart rate data was collected for each participant at a 1 Hz frequency using a chest strap device installed at the base of the sternum. Data was transferred to Polar Beat software in real-time, which was later extracted for data segmentation, where it was segmented for each experimental trial. The raw EDA data collected using an E4 emphatic wristband were pre-processed using segmentation and downsampling procedures. Raw data collected from participants were segmented based on the timestamp for each experimental trial, separating eye-tracking tasks and non-stress/stress activity. After segmentation, EDA data were downsampled from 4 Hz to 1 Hz using Ledalab (V3.4.9), a MATLAB-based skin conductance data analyzer (Benedek & Kaernbach, 2010). Downsampling of EDA data was conducted to match the sampling frequency of heart rate (1 Hz) and to prevent overestimating the significance of the statistical analysis. As EDA data may contain artifacts and noise that can obscure the signals of interest, the artifact removal process was conducted on downsampled data. This study applied a low-pass Butterworth filter to the raw EDA data for artifact removal and adaptive smoothening to remove unwanted noise.
Ledalab, an open-source tool that provides an artifact removal function, filtrated raw EDA data. For the decomposition of raw EDA data in its phasic (EDR) and tonic (EDL) components, a convex-optimizationbased EDA model (cvxEDA) was used (Greco et al., 2016). To process the time-domain feature of HRV, RR-interval data in milliseconds for each participant were extracted from the collected HR data. For HRV analysis, the RR interval data were cleaned by ectopic peak replacement using linear interpolation to produce NN interval data via the Malik method (Malik, 1996;Peltola, 2012). Further time-domain and frequency-domain features of HRV were extracted using a Python package developed by Aurahealthcare (Champseix, 2021). Root mean squares of consecutive differences in adjacent NN (RMSSD) parameters of interest were collected from each participant for all experimental trials.
Participants' physiological parameters (HR, RMSSD, EDR), near-miss recognition performance indicators (FD and TFF), and subjective RPE score data collected during the experiment were imported into the IBM statistical tool to evaluate the hypotheses. Analysis was initiated by verifying that the data met the assumptions for appropriate statistical methods. As participants' data were collected and measured on a continuous scale, scatterplots were utilized to visualize and identify any outliers. Shapiro-Wilk test of normality was conducted to evaluate the data distribution for all participants in both experimental data sets. Shapiro-Wilk returned a p-value > 0.05 for all parameters, showcasing normal distribution except for EDR parameters. Therefore, the dependent t-test and Wilcoxon-Singled Rank test were performed to establish the differences between the two experimental trial categories. Personality traits data collection employs the Likert scale to gauge individuals' subjective evaluations of various personality factors. Hence, to test the internal consistency of the personality questionnaire, Cronbach alpha reliability analysis was utilized. The Cronbach's alpha coefficients for all personality parameters surpassed the widely recommended acceptable threshold (i.e., a coefficient equal to or greater than 0.70), indicating the internal consistency of the personality survey. Additionally, Pearson's correlation was conducted between (NMRP vs FD and TFF) to establish the usability of eye-tracking matrices as near-miss recognition performance indicators and between REP scores and HR to detect the presence of overexertion induced by physical exertion. The results of the statistical analysis are further explained in the sections below.
This section of results elicits the near-miss recognition performance of participants during non-stress and physical stress (overexertion) conditions. To establish the presence of overexertion in participants during physical stress conditions, their HR data over time during physical exertion activity revealed that most participants engaged in moderate-to-high intensity physical activity, and the average recorded RPE score at the end of physical exertion activity was 14 Hard (heavy) level of exertion. This was further verified using the APMHR standard formula. The use of the RPE scale to record participants' exertion rating and their perceived physical exertion was verified, and a Pearson correlation analysis showcased a statistically positive correlation between participants' HR and RPE rating, r (210) = 0.476, p = 0.001<0.01.
Similarly, a second correlation analysis was conducted to verify the usability of fixation matrices as indicators of near-miss recognition performance between NMRP and fixation-derived matrices of fixation duration (FD) and time to first fixation (TFF). Results from the Pearson correlation test showed a statistically significant positive correlation between participants' FD [r (70) = 0.762, p = 0.001<0.01] and TFF [r (70) = 0.608, p = 0.001<0.01] with their NMRP. Figure 7 shows the plot of NMRP and experimental condition across all participants as they performed near-miss recognition activity. Figure 8: Overall average FD and TFF for experimental conditions (Non-stress & Physical-stress)
Furthermore, to evaluate the impact of physical stress (physical exertion) on participants' near-miss recognition performance, their FD and TFF were measured across all AOI from the stimuli images for non-stress and physical stress conditions. With a within-subject experimental design and data manifesting normal distribution, a paired sample t-test was conducted to investigate the impact of physical stress on participants' FD and TFF (table 1). The statistical analysis showed a statistically significant difference between FDs for non-stress and physical stress conditions [z = 6.147, p < 0.001]. On average, there was a 39.43% reduction in participants' FD after they performed a physical exertion trial. Similarly, the paired sample t-test between TFF for non-stress and physical stress trials showed a statistically significant difference between TFFs for non-stress and physical stress conditions [z = 3.707, p < 0.001]. Although there was statistical significance and participants' TFF reduced by 27.37% post-exposure to a physical stress condition, the result was in contrast with the expectation that average TFF would increase as participants' cognitive levels reduced due to increased physical loading. Figure 8 compares the average FD and TFF for both experimental conditions. Therefore, hypothesis H1 is not rejected; the results suggest that induction of overexertion due to moderate-to-high level of physical exertion negatively influences near-miss recognition performance.
Condition Mean Test Statistics (Z) Significance (p-value) FD Non-Stress 2.092 -vs. Physical Stress 1.267 6.147 <0.001* TFF Non-Stress 2.064 -vs. Physical Stress 1.499 3.707 <0.001* Note: t-value statistics from Paired Sample t-test, *p-value<0.05 Table 2: Impact assessment of physical stress on participants' fixation duration (FD) and time to first fixation (TFF) measurement across both conditions
recognition. Statistical analysis investigated the association of EDR, HR, and RMSSD physiological parameters with physical exertion during near-miss recognition activity. The non-stress and physical stress EDR, HR, and RMSSD during the eye-tracking activity were tested to confirm that they satisfy the assumptions of a dependent t-test. These variables were measured on a continuous scale, and the normality test returned p > 0.05 for HR and RMSSD, whereas EDR returned p < 0.05. Therefore, the Paired sample t-test and the Wilcoxon signed rank test were adopted for pairwise comparison of physiological parameters across conditions. Table 2, the statistical test result showed a statistically significant increase in the EDR value of participants during near-miss recognition activity post-physical exertion, z = -3.243, p = 0.001; therefore, we reject hypothesis H2a there is about 51% increase in EDR response due to external stressor of physical exertion.
Similarly, the Paired sample t-test between HR for non-stress and physical stress conditions showed a statistically significant increase in the HR of participants during near-miss recognition activity postphysical exertion, t = -5.036, p = <0.001; therefore, we fail to reject hypothesis H2b as there is an increase in average HR of participants post physical exertion condition. In the case of RMSSD, the time-domain feature of HRV statistically significant mild reduction was observed between RMSSD for non-stress and Physical stress near-miss recognition activity post-physical exertion, t = 3.038, p = 0.005; therefore, we reject hypothesis H2c as the induction of overexertion due to moderate-to-high level of physical exertion negatively impacted the average RMSSD value for participants for near-miss recognition performance. Figure 9 and 10 showcase participants' EDR, HR, and RMSSD plotted across non-stress and physical stress conditions for total (figure 9) and segmented eye-tracking activity (figure 10). In addition, figure 11 showcase means of EDR, HR, and RMSSD across non-stress and physical stress condition. To explore the influence of participants' extraversion (Ex) and neuroticism (Nu) personality traits, nearmiss recognition performance for pre and post-physical exertion eye-tracking activity was evaluated using FD and TFF. Influence assessment was accomplished by grouping participants' personality scores. The Big 5 Personality traits of Ex and Nu were grouped as follows: Below the 25th percentile will be equated to the "low" personality trait group, between the 25th percentile and the 75th percentile would serve as "moderate," and higher than the 75th percentile would be designated as "high." personality trait group. For analysis in this study, "low" and "high" levels of personality groups were selected with low Extraversion (n=10), high Extraversion (n=11), low neuroticism (n=10), and high neuroticism (n=10) participants in each group. Shapiro-Wilk normality test was used to evaluate the distribution of FD and TFF for low and high groups across pre-and post-physical exertion. It returned p > 0.05 and showed a normal distribution. Therefore, Paired sample t-test was utilized to compare the near-miss recognition performance of participants with low and high personality levels across experimental conditions. FD, which represents the visual attention of individuals over AOI, presented statistically significant differences for participants with low Ex (t = 3.722, p < 0.001), high Ex (t = 3.261, p = 0.001), low Nu (t = 4.761, p < 0.001), and high Nu (t = 3.766, p < 0.001) personality traits between pre and post-physical exertion near-miss recognition activity; therefore, we reject hypothesis H3a and H3b (table 3). The percent reduction in FD during post-physical exertion near-miss recognition performance was highest for Low_Nu (48.69%), low_Ex (43.82%) being second, followed by High_Nu (42.02%) and High_Ex (35.68%) represented in figure 12.
Similarly, time to first fixation (TFF), the time the observer takes to fixate on AOI for the first time, indicates adaptability to external stressors. Results show a statistically significant difference for the low Nu personality group between pre and post-physical exertion near-miss recognition activity, t = 2.526, p < 0.013). On average, low Nu group participants' TFF was reduced by 36.85% for near-miss recognition performance post-physical exertion compared to non-stress conditions. Other levels of Ex and Nu present statistically insignificant differences for TFF between pre-and post-exertion near-miss recognition performance, indicating a weak influence of personality trait levels on TTF compared to FD (figure 13). Figure 13: Non-stress vs. Physical stress comparison for time to first fixation (TFF)
In this study, wearable physiological sensors, the Borg scale, and the Big 5 personality survey were utilized to evaluate the impact of overexertion due to moderate-to-high level physical exertion on participants' nearmiss recognition performance. Results show that wearable sensors (Polar H10) could assess the presence of physical exertion as there was a significant correlation between participants' ratings of perceived exertion (moderate-to-high) and HR. The viability of eye-tracking technology to investigate near-miss recognition was evaluated using subjective identification of NMRP score and eye-tracking matrices, resulting in significant correlation. The impact of physical exertion during the physical stress trial reduced participants' near-miss recognition performance (FD & TFF), as there was a statistically significant difference between non-stress and physical stress conditions. These results align with previous research conducted to explore the role of fatigue in situational awareness to recognize hazards in the workplace, reviling reduction in situational awareness as fatigue increases (Namian et al., 2021;Perlman et al., 2014;Uddin et al., 2020).
The mean FD was reduced by 39.43% when participants performed near-miss recognition activity post-physical exertion compared to the non-stress control trial, while TFF decreased by 27.37%. Based on the characteristics of TFF, the expected trend was to showcase an increase in TFF as participants experiencing overexertion would lead to a reduction in cognitive alertness and an increase in their time to fixate on near-miss AOI. This anomaly in the findings can be due to participants having enough experience to nullify the impact of physical exertion and focus on finding near-miss scenarios. In addition, 8 participants reported they had direct or indirect exposure to past injuries that can increase their vigilance post stress condition increasing their ability to identify near-miss scenarios. Third, most of the participants went through their last safety training recently providing them with better vigilance compared to other participants in past research. Thus, reduction in FD and increase in TFF provides justification to including targeted overexertion management protocols is crucial on construction sites to maintain a high level of vigilance, safety behavior, and risk mitigation ability among construction workers.
Physiological data collected in this study using wearable sensors performed well in tracking participants' bio-signals during the experimental trial. The average EDR approximately doubled with an increase of 51% during near-miss recognition activity post physical exertion. EDR, the rapidly changing component of electrodermal activity, responds to external stressors. Therefore, the statistically significant difference between non-stress and physical stress EDR during near-miss recognition activity proves that participants were under physical stress (overexertion), reducing recognition performance. Similarly, it was observed that there is a statistically significant difference between non-stress and physical stress HR and RMSSD measured during near-miss recognition activity. Specifically, participants' average HR increased as the near-miss recognition performance reduced. The root mean square of successive RR interval differences (RMSSD) is the time domain feature widely used predictor of fatigue and strongly associated with hazard recognition performance. Participants' RMSSD value was reduced as their near-miss recognition performance was reduced. This trend suggests that RMSSD can be used as a vital physiological parameter to monitor workers experiencing fatigue due to physical exertion that may lead to deficient nearmiss recognition and reporting ability. Therefore, construction professionals should promote off-the-shelf sensors to track workers' HR, EDR, and RMSSD automatically using the perceived rating of exertion to observe workers' overexertion and related risk. They should also educate workers through a practical training session about fatigue-related risks, encouraging them to identify salient fatigue symptoms and take breaks or stop work when required (Campbell, 2019).
The findings from the influence assessment of personality traits on near-miss recognition performance extend the understanding of personality in safety behavior, specifically how workers with polar levels of personality profile perform near-miss detection under physical exertion. Among the Big 5 personality traits, extraversion and neuroticism were found to rank most highly associated with an individual's fatigue adaptability. Additionally, these two personality traits are closely related to measure variation in individual safety behavior (Conchie, 2013;Golimbet et al., 2007). FD is the visual attention parameter of the eye-tracking matrix; it represents the amount of time spent by an individual observing designated AOI and, therefore, measuring their cognitive load. For FD, participants with low and high extraversion and neurotic personality underperformed during near-miss recognition activity post-exposure to physical exertion activity.
Additionally, individuals with lower neurotic attributes showcased the highest reduction in near-miss recognition performance when exposed to physical exertion. A plausible explanation may lie in the increased exposure to physical stress, as individuals with low neurotic personality traits are usually calm and resilient. However, physiological manipulation due to overexertion induced by physical exertion increases the need to exercise resilience, shown by a significant increase in heart rate measurement due to activating the SNS response, triggering their flight or fight response, consuming mental resources, and reducing cognitive ability. The time to first fixation (TFF) eye-tracking parameter represents observing stimuli before fixating on AOI. This indicates an individual's adaptability to observe the target area as early as possible. Among personality traits, individuals with low levels of neurotic personality showcased statistical significance differences with a reduction in TFF post-physical exertion. This observation can be viewed as an anomaly in this study. Theoretically, TFF increases as participants' ability to fixate on nearmiss AOI increases, taking them longer to isolate the target area due to overexertion caused by physical exertion. Therefore, using TFF as an eye-tracking parameter to evaluate near-miss recognition performance should be reconsidered in future studies. The results demonstrated a relationship between low and high extraversion and neurotic personality with near-miss recognition ability; employers will benefit from evaluating these personalities before recruitment and deployment in the field of new hires. Considering the negative effect of low neurotic and extraversion personality traits on safety behavior, in addition to relocating such individuals, one possible intervention effort is to enhance exertion management through supervisors' intervention because this approach can provide employers the ability to manage the negative effect of extraversion and neurotic personality individuals on near-miss identification and reporting.
Physical exertion manifested as physical stress in the construction industry could reduce workers' safety performance and near-miss identification ability by increasing unsafe behavior on job sites. This study investigates the impact of overexertion induced by physical exertion on workers' near-miss recognition performance using eye-tracking technology, physiological parameters, and personality indicators. The results confirm that overexertion induced by physical exertion impaired participants' ability to identify and perceive near-misses, which may increase at-risk safety behavior and vigilance at work. The strong correlation between the RPE Borg scale and HR suggests the usability of the RPE scale as a valid measure of physical overexertion, as previous studies suggested. A similar significant correlation between NMRP% and eye-tracking matrices proves that eye-tracking technology can objectively measure near-miss recognition performance. The impact assessment shows that workers' near-miss recognition performance reduced post-exposure to physical exertion by 39.43% for fixation duration and 27.37% for time to first fixation. Statistically significant differences in physiological parameters of HR, EDR, and RMSSD during near-miss recognition activity pre and post-physical exertion suggest the inclusion of wearable sensors as a real-time assessment tool for exertion intervention to limit inattentional blindness towards near-misses. Lastly, results of personality level assessment for extraversion and neuroticism traits on near-miss recognition performance pre and post-physical exertion showcase that individuals with low neuroticism and extraversion personality levels exhibit a higher reduction in recognition performance when overexerted.
Despite the findings, some limitations remain to be acknowledged for future directives. First, the current study utilized static real-world construction fatal four stimuli images to measure near-miss recognition performance. Future research could explore dynamic scenarios using state-of-the-art eyetracking technology and immersive virtual reality that can provide more realistic near-miss incidents and more robust experimental evidence. Second, the participant sample size recruited for this study met the requirements for experimental design for a non-clinical experiment. However, it should be noted that these participants consisted of African American, Hispanic, or Latino, Middle Eastern, and Caucasian ethnicities with male and female genders. Hispanic construction workers usually experience a curtailed safety climate, making them more vulnerable to risk-taking behavior (Marin et al., 2015). Future studies could explore diverse ethnicities of the South and Central American populations, providing larger sample sizes to generalize to a broader working population in the United States. Third, the present study exposed participants to physical exertion using manual material handling activity. However, future studies could utilize more realistic construction activity and additional confounding factors such as environmental variation, prior fatigue, and complex tasks for inducing physical exertion (e.g., scaffold installation tasks, including picking, carrying, and assembly with exposure to heat, humidity, and physical stress by varying times to induce fatigue). Fourth, the TFF (Time to First Fixation) matrix in the eye-tracking data exhibited anomalous behavior that showcase needs for further investigation. Future studies should consider evaluating additional confounding factors, such as pre-experiment physical exertion, sleep cycles, and fatigue levels, to assess their impact on construction-specific reaction tasks, particularly in near-miss recognition activities.