Predicting Occupant Annoyance in Acoustic-Thermal Compound Environments
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis is a well-designed and timely study that bridges environmental psychology, physiology, and machine learning in a study context. The integration of physiological signals with environmental parameters for modeling annoyance is particularly commendable. My comments regarding the manuscript are as follow:
- Some sentences throughout the manuscript are grammatically awkward or overly complex. A professional language edit is recommended.
- While the setup is described in text, a more detailed schematic (framework) with step by step description and labeled sensors and placement zones (with connection to input and output) would help readers understand the research and replicate the experiment.
- While the Random Forest model performed well, a more detailed description and rationale for choosing this model over other ML techniques is needed to be added Section 5.
- The study acknowledges its male-only sample. Why is this the case? Please consider elaborating more on how this may limit generalizability and suggest avenues for broader demographic inclusion, and perhaps adding this as a big limitation on interpreting the findings.
Author Response
Comments 1: Some sentences throughout the manuscript are grammatically awkward or overly complex. A professional language edit is recommended.
Response 1: We have touched up the language.
Comments 2: While the setup is described in text, a more detailed schematic (framework) with step by step description and labeled sensors and placement zones (with connection to input and output) would help readers understand the research and replicate the experiment.
Response 2: We have added the standards and diagrams for each point in Chapter 2.
Comments 3: While the Random Forest model performed well, a more detailed description and rationale for choosing this model over other ML techniques is needed to be added Section 5.
Response 3: 3. A detailed description and advantages of the random forest model have been added to chapter V, lines 453-471.
(1) There are non-linear relationships and complex interactions between the screened predictor variables (PMV value, roughness, PNS/SNS index, skin temperature), and the Random Forest model is able to automatically capture the non-linear relationships and interaction effects between the variables by integrating multiple decision trees without the need for manually designing feature combinations.
(2) There are dynamic environmental noise (e.g., sound pressure level fluctuations caused by compressor start/stop) and individual differences (e.g., differences in physiological responses of subjects) in the experimental data, which may lead to data noise and outliers. The random forest model reduces variance through Bagging (self-sampling), single trees are insensitive to noise, and the model is robust after integration.
(3) The sample size is limited (60 sets of total data), requiring an efficient and stable algorithm. Random forest model has few hyperparameters (e.g., number of trees, maximum depth), fast training speed, and stable results in multiple runs.
(4) Overfitting needs to be avoided in small sample scenarios. Random forest model reduces single tree correlation by randomly selecting features and samples, and has strong generalization ability after integration.
Comments 4: The study acknowledges its male-only sample. Why is this the case? Please consider elaborating more on how this may limit generalizability and suggest avenues for broader demographic inclusion, and perhaps adding this as a big limitation on interpreting the findings.
Response 4: The selection of a single-sex sample for this study is mainly based on the following considerations:
(1) Demographic characteristics of the industry: according to the China Federation of Logistics and Purchasing (CFLP)‘s “Survey Report on Truck Drivers” Employment Situation in 2021’, over 90% of long-haul truck drivers are male, constituting the absolute main body of the profession. The gender distribution of the samples in this paper is highly compatible with the actual gender structure of the target group, which is in line with the basic principle of ‘priority of sample representativeness’ in social science research.
(2) The need for experimental control: It has been shown that gender differences may affect thermal sensitivity (e.g., women are more sensitive to temperature fluctuations) and autonomic nervous system response characteristics (e.g., sympathetic/parasympathetic balance) (Karjalainen, 2012; Chaudhuri et al., 2018). In preliminary explorations of environmental-physiological interaction mechanisms, controlling for sex variables has helped to reduce confounding effects.
We fully acknowledge that this design can limit the generalizability of findings. For example, sex-related differences in skin thermoregulation (e.g., lower basal metabolic rate in women), patterns of heart rate variability (e.g., higher parasympathetic activity in women under heat stress), and differences in subjective annoyance thresholds may lead to variations in the predictive performance of the model in other populations. In addition, age factors (e.g., decreased thermal regulation in older adults) were not included and may also affect extrapolation of conclusions. To address this issue, we have added a new paragraph in Chapter 6, ‘Limitations and future work’, which emphasizes the need for gender-balanced and age-diverse validation, including:
- Expanding the test to include female drivers and middle-aged and older age groups;
- Testing the robustness of the model to a sex/age-specific physiological baseline;
- Introducing demographic characteristics (e.g. BMI, metabolic rate) as auxiliary predictor variables.
In particular, it should be noted that the current model is still representative and engineering guidance for the primary target population (male drivers).
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsDear authors,
The following are my remarks regarding the manuscript:
- Abstract: It would be nice to see some numerical data from the results obtained.
- After reading the Abstract and the first paragraph of the Introduction I got confused. They seem to be from two different scenarios. I suggest the authors either to include in the Abstract where the measurements took place or start the Introduction in a different manner.
- Line 32: “The use of parking air conditioning to regulate the temperature inside the cockpit has the advantage of energy saving and convenience”. How can the use of air conditioning have the advantage of energy savings? Maybe to create comfort.
- The Introduction section needs serious improvements. Authors should provide more context for the research, highlight the identified gap, novelty and necessity of the study.
- Line 132: “The subjects entered the cabin, started the air-conditioner and set the temperature to 24°C. They then proceeded to drive for 15 minutes at four different working conditions.” It is not clear when the measurements occur. When did the subjects drive or when they were in the car park?
- Abbreviations should be used after the whole written words. There are a lot of abbreviations used and are not explained (SNS, MeanRR, etc.)
- Statistical analysis: Pearson correlation is used for quantitative data. For quantitative and qualitative data Kendall and/or Spearman coefficient should be used.
- Line 208 and Figure 4a: PMV range is between -3 and +3. Why did the authors present in Figure 4a the PMV from +4?
- Line 209: “observed that the decreasing trend of condition 1 and condition 2 is more pronounced”. What are these conditions? Authors should explain them
- Table 5, Table 8 - Usually, in the scientific literature where statistical analysis is used, the * indicates a p value <0.05 and ** indicates a p value <0.01. Authors should explain what the ** represents near Table 8.
- I suggest the authors add a Discussion section and Limitations of the study.
- For a research study, the Reference section is weak and should be improved and updated.
Author Response
Comments 1: Abstract: It would be nice to see some numerical data from the results obtained.
Response 1: We have added specific data to the abstract.
Comments 2: After reading the Abstract and the first paragraph of the Introduction I got confused. They seem to be from two different scenarios. I suggest the authors either to include in the Abstract where the measurements took place or start the Introduction in a different manner.
Response 2: We have stated the location of the measurements in the abstract and have adjusted the way the introduction begins, see lines 15-37, 42-55.
Comments 3: Line 32: “The use of parking air conditioning to regulate the temperature inside the cockpit has the advantage of energy saving and convenience”. How can the use of air conditioning have the advantage of energy savings? Maybe to create comfort.
Response 3: We have added the advantages of parking air conditioning to the text, see lines 45-48.
Comments 4: The Introduction section needs serious improvements. Authors should provide more context for the research, highlight the identified gap, novelty and necessity of the study.
Response 4: We have improved the introduction section by adding the context of the study, the research gap, the innovative type and the need, see lines 61-78, 91-105.
Comments 5: Line 132: “The subjects entered the cabin, started the air-conditioner and set the temperature to 24°C. They then proceeded to drive for 15 minutes at four different working conditions.” It is not clear when the measurements occur. When did the subjects drive or when they were in the car park?
Response 5: We have added to the paper, see lines 143-152.
The test was carried out in a simulated cab to simulate the reality of a lorry parked in a summer environment. Once the subject was seated in the cab, the air conditioning was activated and set to one of the cooling modes, with the temperature set to 24°C. Measurements were taken at the same time as the air conditioning was activated, and were completed after 15 minutes. The parked air conditioner in this test had four cooling modes, so the subject had to repeat the test more than four times. At the end of each mode, the thermal and acoustic environments were restored to their initial state, and the next mode was measured.
Comments 6: Abbreviations should be used after the whole written words. There are a lot of abbreviations used and are not explained (SNS, MeanRR, etc.)
Response 6: We have added full names to the abbreviations and added a new Table 5 explaining the meaning of the parameters.
Comments 7: Statistical analysis: Pearson correlation is used for quantitative data. For quantitative and qualitative data Kendall and/or Spearman coefficient should be used.
Response 7: In this study, Pearson correlation coefficients were used in analyzing the correlation between environmental parameters (e.g., PMV value, roughness R) and subjective annoyance (5-level scale) for the following reasons:
(1) Data type suitability: although the annoyance vote is an ordered categorical variable (5-level scale), the nature of its assignment is an equidistant scale (0-4 points), and the normality of the scale scores was verified by the pre-experiment (Shapiro-Wilk test, p>0.05), which meets the application prerequisite of the Pearson correlation coefficient.
(2) Validation of methodological comparison: in response to the review suggestions, we additionally calculated the Spearman and Kendall correlation coefficients (as shown in Table 6). The results showed that the Pearson coefficient of PMV and annoyance (r=0.931) was highly consistent with Spearman (ρ=0.918) and Kendall (τ=0.802) and had the same level of significance (p<0.01), indicating the robustness of the analysis.
Comments 8: Line 208 and Figure 4a: PMV range is between -3 and +3. Why did the authors present in Figure 4a the PMV from +4?
Response 8: The PMV values in this paper are calculated according to the formula of the Fanger model (Equation 1). The design range of PMV is indeed from -3 to +3, which corresponds to a 7-level thermal sensory vote. However, the PMV value may exceed the upper limit when facing extreme environmental parameters in the heat. Meanwhile, the mathematical expression of PMV does not have a strict upper limit in theory, and its ‘-3 to +3’ is a statistical fitting range based on thermal sensory polling. Therefore an extension is made to facilitate the description of sensations in extreme environments.
Comments 9: Line 209: “observed that the decreasing trend of condition 1 and condition 2 is more pronounced”. What are these conditions? Authors should explain them
Response 9: These conditions refer specifically to the four cooling modes of the parking air conditioner, each of which is a condition, and the four modes correspond to four noise levels and cooling capacities. This has been explained in the text, see lines 217-219.
Comments 10: Table 5, Table 8 - Usually, in the scientific literature where statistical analysis is used, the * indicates a p value <0.05 and ** indicates a p value <0.01. Authors should explain what the ** represents near Table 8.
Response 10: We have added the note in Table 8: **p<=0.01.
Comments 11: I suggest the authors add a Discussion section and Limitations of the study.
Response 11: We have added a new paragraph, ‘Discussion and limitations’, to chapter VI.
Comments 12: For a research study, the Reference section is weak and should be improved and updated.
Response 12: We have improved the references.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsOn page 2, lines 43-47, the authors correctly identify that "a substantial body of current literature has been dedicated to the analysis of building rooms or outdoor environments, with a paucity of studies conducted on acoustic-thermal composite environments in vehicle driver's cockpits." However, the distinctiveness and necessity of this research in addressing this research gap could be more explicitly articulated.
Methodology
The experimental procedures and measurement methods are thoroughly described, but the methodology could be strengthened by providing clearer details on subject selection criteria and ethical approval procedures. While the current description comprehensively describes the measurement protocol, it lacks sufficient transparency about the participant recruitment strategy and ethical considerations, which are important factors in methodological rigour and reproducibility.
Data
The statistical analyses were conducted appropriately, and the correlation heatmap in Figure 8 effectively illustrates the relationships between variables. However, the reproducibility of this study could be improved by providing additional details about the data preprocessing procedures, particularly the methods used for outlier removal and data normalisation during the analysis phase.
Results.
The results of the study were logically interpreted and made important findings, most notably that thermal environments have a greater impact on annoyance than acoustic environments. Nevertheless, a more robust discussion comparing these findings to existing literature would better contextualise the results in a broader research setting.
On page 12, lines 319-330, when discussing the prediction model utilizing the Random Forest algorithm, the authors state: "The accuracy of the model prediction is estimated by the 10-fold cross-validation method, and the coefficient of determination (R²), the mean absolute error (MAE) and the root mean square error (RMSE) are selected as the evaluation indexes..." However, without comparative analysis against alternative machine learning algorithms, it remains difficult to ascertain whether Random Forest represents the optimal choice for this application.
Regarding Figure 11 described on page 13, lines 336-354, the authors note: "Thus the three-dimensional relationship model of PA, R and PMV was constructed as shown in Figure 11." Unfortunately, specific guidelines for practical interpretation and application of this three-dimensional model are insufficient.
In the conclusion on page 14, lines 358-362, the authors summarize: "The impact of varying acoustic and thermal environments on human physiological parameters and subjective annoyance when environmental changes in the cabin were examined." However, the discussion of research limitations and future research directions requires expansion.
Limitations
The experiment was conducted with only 20 adult males, so there was a lack of gender diversity. A more diverse pool of participants with different demographics is needed to generalise the findings.
The experiment was conducted in a simulated environment (a 2×2×2 m simulated space) rather than in a real truck cab, which may limit the applicability of the findings to real-world driving environments.
The correlation analysis was well done, but further analysis, such as structural equation modelling (SEM), could provide clearer insight into the causal relationships between variables.
As this study only analysed the short-term effects of 15-minute intervals, further research is needed on the patterns of annoyance adaptation during longer exposures.
Author Response
Comments 1: On page 2, lines 43-47, the authors correctly identify that "a substantial body of current literature has been dedicated to the analysis of building rooms or outdoor environments, with a paucity of studies conducted on acoustic-thermal composite environments in vehicle driver's cockpits." However, the distinctiveness and necessity of this research in addressing this research gap could be more explicitly articulated.
Response 1: We have improved the introduction section by adding the context of the study, the research gap, the innovative type and the need, see lines 61-78, 91-105.
Comments 2: The experimental procedures and measurement methods are thoroughly described, but the methodology could be strengthened by providing clearer details on subject selection criteria and ethical approval procedures. While the current description comprehensively describes the measurement protocol, it lacks sufficient transparency about the participant recruitment strategy and ethical considerations, which are important factors in methodological rigour and reproducibility.
Response 2: We have made changes to 2.1Human subjects to address this issue.
This study strictly followed the ethical guidelines of the Declaration of Helsinki and was reviewed by the Ethics Committee of Wuhan University of Science and Technology during the experimental design phase. The proportion of males among lorry driver practi-tioners is more than 90%. High-temperature exposure experiments need to consider physiological adaptability thresholds. This study was approved by the Ethics Committee. Subjects were recruited using a tiered screening mechanism. Firstly, 20 healthy male prac-titioners were screened by an occupational adaptability assessment. Their anthropometric information is shown in Table 1. The screening process was meticulous in its adherence to the following ethical protection measures:
(1) Multi-dimensional health screening: a three-level medical assessment system was adopted, which included a basic questionnaire survey (history of cardiovascu-lar/respiratory/skin diseases), a second-level clinical examination (resting electrocardio-gram, skin sensitivity test), and a third-level high-temperature acclimatization pre-experiment (30-minute tolerance test in a 40°C environment), to ensure that the physi-ological conditions of the subjects were in line with the experimental safety standards.
(2) Dynamic informed consent: a written consent form is signed prior to the experi-ment, covering experimental risks, data use and voluntary withdrawal mechanisms, and the right to information is reiterated through concise prompts at each stage.
(3) Long-term protection system: establish a database of subjects' health records and provide 3-year tracking and follow-up services after the experiment, including quarterly cardiovascular function assessment and skin health monitoring. At the same time, a spe-cial protection fund will be set up to provide medical support for possible delayed health effects, and encrypted storage and controlled access to experimental data will be realized through blockchain technology.
The research team would like to make a special note: although the gender of the sub-jects is limited for safety reasons at this stage, a pre-study on the physiological character-istics of female drivers has been carried out at the same time, so as to lay the foundation for the subsequent research on gender-differentiated protection. It is imperative to note that all subjects possess the inherent right to withdraw from participation at any time and to receive an experimental allowance that meets international standards. This is in order to ensure the rights and welfare of the subjects.
Comments 3: The statistical analyses were conducted appropriately, and the correlation heatmap in Figure 8 effectively illustrates the relationships between variables. However, the reproducibility of this study could be improved by providing additional details about the data preprocessing procedures, particularly the methods used for outlier removal and data normalisation during the analysis phase.
Response 3: We have added a new subsection 3.1Data preprocessing in Chapter 3, see lines 262-283.
Comments 4: The results of the study were logically interpreted and made important findings, most notably that thermal environments have a greater impact on annoyance than acoustic environments. Nevertheless, a more robust discussion comparing these findings to existing literature would better contextualise the results in a broader research setting.
Response 4: We have added a discussion of the conclusions of this paper in comparison with the existing literature in Chapter 6, see lines 552-558.
Comments 5: On page 12, lines 319-330, when discussing the prediction model utilizing the Random Forest algorithm, the authors state: "The accuracy of the model prediction is estimated by the 10-fold cross-validation method, and the coefficient of determination (R²), the mean absolute error (MAE) and the root mean square error (RMSE) are selected as the evaluation indexes..." However, without comparative analysis against alternative machine learning algorithms, it remains difficult to ascertain whether Random Forest represents the optimal choice for this application.
Response 5: We have added the advantages of the Random Forest algorithm over other algorithms, see lines 453-471.
(1) There are nonlinear relationships and complex interactions between the screened predictor variables (PMV value, roughness, PNS/SNS index, skin temperature), and the Random Forest model, by integrating multiple decision trees, is able to automatically capture the nonlinear relationships and interaction effects between the variables without the need for manually designing the feature combinations.
(2) There are dynamic environmental noise (e.g., sound pressure level fluctuations caused by compressor start/stop) and individual differences (e.g., differences in physiological responses of subjects) in the experimental data, which may lead to data noise and outliers. The random forest model reduces variance through Bagging (self-sampling), single trees are insensitive to noise, and the model is robust after integration.
3) The sample size is limited (60 sets of total data), requiring an efficient and stable algorithm. Random forest model has few hyperparameters (e.g., number of trees, maximum depth), fast training speed, and stable results in multiple runs.
4) Overfitting needs to be avoided in small sample scenarios. Random forest model reduces single tree correlation by randomly selecting features and samples, and has strong generalisation ability after integration.
Comments 6: Regarding Figure 11 described on page 13, lines 336-354, the authors note: "Thus the three-dimensional relationship model of PA, R and PMV was constructed as shown in Figure 11." Unfortunately, specific guidelines for practical interpretation and application of this three-dimensional model are insufficient.
Response 6: We have added practical explanations and guidance on the application of this 3D model in Chapter 5, see lines 508-525.
Comments 7: The experiment was conducted with only 20 adult males, so there was a lack of gender diversity. A more diverse pool of participants with different demographics is needed to generalise the findings.
Response 7: We have added research limitations and future work to Chapter 6, see lines 559-574.
Comments 8: The experiment was conducted in a simulated environment (a 2×2×2 m simulated space) rather than in a real truck cab, which may limit the applicability of the findings to real-world driving environments.
Response 8: We fully agree with the reviewing experts on the importance of the validation of real driving scenarios, and the choice of conducting tests in a 2 x 2 x 2 m simulated cab in this study is mainly based on the following scientific considerations:
(1) Precise control of variables: Real truck cabs are significantly disturbed by external meteorological conditions (e.g., sunlight radiation, fluctuations in ambient temperature and humidity) and dynamic traffic noise (e.g., road vibration, engine transient conditions), making it difficult to stably reproduce extreme boundary conditions. The simulation environment can precisely regulate the acoustic and thermal parameters to ensure that the experimental conditions meet the needs of theoretical model verification.
(2) Experimental reproducibility: The international standards ISO 14505-1 and ASHRAE 55 both recommend the use of controlled laboratory environments for thermal comfort studies, in order to eliminate the effects of individual behaviours (e.g., opening and closing of windows, seat adjustments) and external perturbations on the consistency of data. In this study, high reproducibility of data across subjects and operating conditions was achieved by standardizing the layout of measurement points and working condition reset mechanisms (e.g., electric heating/acoustic panels restoring the initial state).
(3) Safety and Ethical Compliance: High temperature-noise exposure experiments require strict protection of subject safety. The real cab is limited by the confined space and emergency response time, which may increase the risk of heat stress. The simulated environment is equipped with real-time physiological monitoring and an emergency cooling system (response time <10 seconds), which is in full compliance with the Declaration of Helsinki on the safety of high-risk experiments.
Comments 9: As this study only analysed the short-term effects of 15-minute intervals, further research is needed on the patterns of annoyance adaptation during longer exposures.
Response 9: We have already stated in our outlook that we will carry out further research on longer-term forecasts of the degree of annoyance in the future.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsOverview
The manuscript investigates how simultaneous acoustic and thermal stressors inside a stationary truck‑cab mock‑up affect occupants’ subjective annoyance and related physiological responses. Twenty young male volunteers sat through four air‑conditioner settings that coupled decreasing temperature with increasing noise; each minute, the authors logged environmental variables, psycho‑acoustic metrics, skin temperature, heart‑rate‑variability indices and a five‑point annoyance rating. Using correlation screening, they selected PMV, acoustic roughness, mean skin temperature, and composite parasympathetic and sympathetic HRV indices as predictors and trained a random‑forest model that explained about 94 % of the variance in average annoyance. They report that thermal factors dominate acoustic ones, that physiological markers track annoyance closely, and that their model could inform noise‑reduction and climate‑control strategies for truck cabs.
Relevance to Electronics
The manuscript is remotely relevant to electronics. While the work does use signal‑processing of ECG and acoustic data and applies a random‑forest algorithm, those elements serve as tools rather than as primary contributions to electronics science. A journal on HVAC, human comfort or bioclimate might be more appropriate
Originality
There are prior multimodal comfort studies that combine acoustic and thermal stimuli have focused on offices, dwellings and passenger cars. The present work’s main novel aspect is its focus on a stationary truck cab fitted with a roof‑mounted parking air‑conditioner, coupled with the attempt to fuse psycho‑acoustic metrics, skin temperature and ultra‑short HRV composites into a data‑driven annoyance model. That specific vehicle context and the use of a five‑variable random‑forest regressor have not, to my knowledge, appeared together in the literature. However, the underlying experimental paradigm closely mirrors prior laboratory studies, and the machine‑learning component represents an incremental substitution for conventional regression rather than a conceptual advance. Consequently, the manuscript offers a modest extension of existing work rather than a step‑change in theory or measurement technique.
Methodology
The experimental platform is a 2 m × 2 m × 2 m mock‑up described only as “similar to the truck cockpit”. No construction drawing, surface‐temperature map or glazing specification is provided, so the radiant field is unknown. The authors explicitly assume mean radiant temperature equals air temperature when computing PMV, thereby eliminating radiant exchange from the heat balance; in small metal‑shelled cabs that simplification can shift PMV by up to one vote. Four test “working conditions” are generated by changing the parking‑air‑conditioner fan speed, which simultaneously lowers temperature and raises noise. Because thermal and acoustic stimuli co‑vary, it is impossible to attribute the reported dominance of thermal factors to temperature alone.
Moreover, the paper gives no indication that the order of the four conditions was randomised or counter‑balanced; adaptation and fatigue effects are therefore confounded with the “time” factor used in the ANOVA. Participant selection is restricted to twenty young men, limiting generalisability and risking a Type‑II error if sex or age modulates annoyance.
Physiological and subjective measurements raise further concerns. Electrocardiograms are sampled at 400 Hz and processed in one‐minute windows, yet frequency‑domain HRV metrics are unreliable at that duration; the authors nevertheless aggregate MeanRR, RMSSD and SD1 into an unvalidated “parasympathetic” index and MeanHR, Stress Index and SD2 into a “sympathetic” index without disclosing the weighting formula. Skin temperature is measured at four ISO 9886 sites, but contact pressure, sweat artefacts, and probe calibration are undocumented. Acoustic data rely on a single microphone whose placement relative to airflow is not described, while typical cab airflow velocities require a more sensitive instrument. Subjective annoyance is collected verbally every minute with a bespoke five‑point scale unanchored to ISO 10551 or ISO 15666; repeated interrogation can induce demand characteristics, and the study reports no test–retest reliability.
Analytical treatment compounds these weaknesses. The repeated‑measures ANOVA treats “working condition” as a between‑subject factor, inflating degrees of freedom; inconsistent p‑values (interaction reported significant despite p = 0.279) suggest further mis‐specification, and multiple comparisons are left uncorrected. Continuous time‑series data are autocorrelated, yet Pearson and Spearman coefficients are calculated as if rows were independent.
For machine learning, the authors first average data across sixteen subjects, collapsing 1 200 observations into sixty and erasing within‑subject variance; they then train a random‑forest model on forty‑eight rows and test on twelve, claiming an R² of 0.94. Because PMV (a composite of air temperature, humidity and speed) is included alongside skin temperature, “feature leakage” inflates apparent accuracy, and cross‑validation folds are not independent No baseline linear model is provided, confidence intervals are absent, and leave‑one‑subject‑out validation is ignored. Collectively these issues undermine the internal validity of the findings and limit their applicability beyond the study sample. ML could ultimately be unnecessary in this context and is poorly implemented in the current methodology.
Manuscript Structure
Ok
Use of English
Use of English is ok. Some sentences are overly long. Verb tenses change, some articles are often missing (“the cockpit noise” becomes “cockpit noise”), and several technical terms are mistranslated (“jitteriness” for fluctuation strength, “annoyance” for “discomfort”).
Figures/Tables
Figs/Tables are readable and can be interpreted.
Bibliography
Ok
Comments on the Quality of English LanguageUse of English is ok. Some sentences are overly long. Verb tenses change, some articles are often missing (“the cockpit noise” becomes “cockpit noise”), and several technical terms are mistranslated (“jitteriness” for fluctuation strength, “annoyance” for “discomfort”).
Author Response
Comments 1: The experimental platform is a 2 m×2 m×2 m mock up described only as“similar to the truck cockpit”. No construction drawing, surface‐temperature map or glazing specification is provided, so the radiant field is unknown. The authors explicitly assume mean radiant temperature equals air temperature when computing PMV, thereby eliminating radiant exchange from the heat balance; in small metal shelled cabs that simplification can shift PMV by up to one vote. Four test “working conditions” are generated by changing the parking air conditioner fan speed, which simultaneously lowers temperature and raises noise. Because thermal and acoustic stimuli co vary, it is impossible to attribute the reported dominance of thermal factors to temperature alone.
Response 1: The interior of the simulated cab is made of blue extruded plastic sheet, a material with low thermal conductivity. Meanwhile, before starting the measurements, the environment in the simulated cab was thermostatically controlled so that the temperature of all surfaces in the room was close to the air temperature, at which point it was reasonable to assume that the average radiant temperature was equal to the air temperature.
The order of the four test ‘conditions’ was balanced and generated by changing the speed of the parked air conditioning fan, an operation that simultaneously reduces temperature and increases noise. Therefore, this paper used a repeated measures ANOVA, which analysed time effects and inter-individual variability for the same subjects experiencing multiple working conditions. A two-factor ANOVA warming exercise was used to analyse the main and interaction effects of changes in acoustic factors and PMV values on subjective annoyance, before finally obtaining the dominance of the thermal factor in the operation of the parked air conditioner in this environment.
Comments 2: Moreover, the paper gives no indication that the order of the four conditions was randomised or counter balanced; adaptation and fatigue effects are therefore confounded with the “time” factor used in the ANOVA. Participant selection is restricted to twenty young men, limiting generalisability and risking a Type II error if sex or age modulates annoyance.
Response 2: The selection of a single-sex sample for this study is mainly based on the following considerations:
(1) Demographic characteristics of the industry: according to the China Federation of Logistics and Purchasing (CFLP)‘s “Survey Report on Truck Drivers” Employment Situation in 2021’, over 90% of long-haul truck drivers are male, constituting the absolute main body of the profession. The gender distribution of the samples in this paper is highly compatible with the actual gender structure of the target group, which is in line with the basic principle of ‘priority of sample representativeness’ in social science research.
(2) The need for experimental control: It has been shown that gender differences may affect thermal sensitivity (e.g., women are more sensitive to temperature fluctuations) and autonomic nervous system response characteristics (e.g., sympathetic/parasympathetic balance) (Karjalainen, 2012; Chaudhuri et al., 2018). In preliminary explorations of environmental-physiological interaction mechanisms, controlling for sex variables has helped to reduce confounding effects.
We fully acknowledge that this design can limit the generalizability of findings. For example, sex-related differences in skin thermoregulation (e.g., lower basal metabolic rate in women), patterns of heart rate variability (e.g., higher parasympathetic activity in women under heat stress), and differences in subjective annoyance thresholds may lead to variations in the predictive performance of the model in other populations. In addition, age factors (e.g., decreased thermal regulation in older adults) were not included and may also affect extrapolation of conclusions. To address this issue, we have added a new paragraph in Chapter 6, ‘Limitations and future work’, which emphasizes the need for gender-balanced and age-diverse validation, including:
- Expanding the test to include female drivers and middle-aged and older age groups;
- Testing the robustness of the model to a sex/age-specific physiological baseline;
- Introducing demographic characteristics (e.g. BMI, metabolic rate) as auxiliary predictor variables.
In particular, it should be noted that the current model is still representative and engineering guidance for the primary target population (male drivers).
Comments 3: Physiological and subjective measurements raise further concerns. Electrocardiograms are sampled at 400 Hz and processed in one‐minute windows, yet frequency domain HRV metrics are unreliable at that duration; the authors nevertheless aggregate MeanRR, RMSSD and SD1 into an unvalidated “parasympathetic” index and MeanHR, Stress Index and SD2 into a “sympathetic” index without disclosing the weighting formula. Skin temperature is measured at four ISO 9886 sites, but contact pressure, sweat artefacts, and probe calibration are undocumented. Acoustic data rely on a single microphone whose placement relative to airflow is not described, while typical cab airflow velocities require a more sensitive instrument. Subjective annoyance is collected verbally every minute with a bespoke five point scale unanchored to ISO 10551 or ISO 15666; repeated interrogation can induce demand characteristics, and the study reports no test–retest reliability.
Response 3: The standard time for calculating HRV index in frequency domain is 5min, but this is applicable to HRV analysis in resting state. This paper belongs to a special scenario where the signal may change rapidly over time, so we need to use segmented analysis, and the segmented time used in this paper is 1min, which can avoid the spectral distortion caused by the non-stationarity of the signal in the long time.The PNS index and SNS index are calculated directly according to the software Kubios HRV Standard, and there is a reference on the calculation of this software here ( Tarvainen M P, Niskanen J P, Lipponen J A, et al. Kubios HRV-heart rate variability analysis software[J]. Computer methods and programmes in biomedicine, 2014, 113(1): 210-220.) The literature has an indication of where these two indices come from.
Comments 4: Physiological and subjective measurements raise further concerns. Electrocardiograms are sampled at 400 Hz and processed in one‐minute windows, yet frequency domain HRV metrics are unreliable at that duration; the authors nevertheless aggregate MeanRR, RMSSD and SD1 into an unvalidated “parasympathetic” index and MeanHR, Stress Index and SD2 into a “sympathetic” index without disclosing the weighting formula. Skin temperature is measured at four ISO 9886 sites, but contact pressure, sweat artefacts, and probe calibration are undocumented. Acoustic data rely on a single microphone whose placement relative to airflow is not described, while typical cab airflow velocities require a more sensitive instrument. Subjective annoyance is collected verbally every minute with a bespoke five point scale unanchored to ISO 10551 or ISO 15666; repeated interrogation can induce demand characteristics, and the study reports no test–retest reliability.
Response 4: The standard time for calculating HRV index in frequency domain is 5min, but this is applicable to HRV analysis in resting state. This paper belongs to a special scenario where the signal may change rapidly over time, so we need to use segmented analysis, and the segmented time used in this paper is 1min, which can avoid the spectral distortion caused by the non-stationarity of the signal in the long time.The PNS index and SNS index are calculated directly according to the software Kubios HRV Standard, and there is a reference on the calculation of this software here ( Tarvainen M P, Niskanen J P, Lipponen J A, et al. Kubios HRV-heart rate variability analysis software[J]. Computer methods and programmes in biomedicine, 2014, 113(1): 210-220.) The literature has an indication of where these two indices come from.
This study strictly followed the ISO 9886:2004 standard for the selection and weight assignment of skin temperature measurement points (4-point weighting scheme). Additional notes for measurement details:
Comments 5: The location of the single microphone is not specified and airflow effects are not considered.
Response 5: (1) Contact pressure control: medical grade silicone patch (AS6221 sensor supporting) is used to fix the probe, the elastic modulus of the patch is 0.5 MPa, and it can adapt to the curvature of the skin surface to ensure that the contact pressure is stable at 5±1 kPa (verified by pre-experiment).
(2) Sweat artefact suppression: Alcohol cleansing of the subject's skin and application of medical coupling agent were performed before the experiment to reduce contact impedance and sweat interference; meanwhile, subjects with abnormal sweat rate (n=2) were screened out in the high-temperature pre-experiment stage (40°C/30 min) to ensure data reliability.
(3) Probe calibration: All sensors were calibrated by a secondary metrology institution (Certificate No.: HB2023-TS001), with a calibration temperature range of 20-50℃ and an error of ±0.1℃. Drift test was carried out before and after the experiment using a constant temperature water bath (37±0.1℃) with a maximum offset of <0.05℃.
Comments 6: Customized scales do not conform to ISO standards, repeatedly ask for or elicit demand characteristics, and do not validate retest reliability.
Response 6: (1) Scale design basis: The present study used a 5-level unipolar scale (Table 3), whose grading logic is compatible with the ISO 10551 Heat Sensation Scale (level 7) and the ISO 15666 Noise Annoyance Scale (level 5), and referred to the validation scheme of Zhang Lu and Guan et al. In order to reduce the cognitive load of the scale, the ‘Neutral’ option was merged into the ‘Slightly Annoyed’ level to meet the need for rapid assessment in dynamic environments.
(2) Demand characteristics control: The experiment was designed as a double-blind experiment, in which subjects did not know the specific order of the working conditions, and the operator only prompted ‘please vote according to your current feelings’, avoiding guiding language.
(3) Retest reliability: 16 subjects were retested at one-week intervals (under the same working conditions), and the Spearman's correlation coefficient ρ=0.86 (p<0.01) and Cronbach's α=0.89 confirmed that the scale had a high degree of temporal stability.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThanks for addressing the comments. No further comments!
Author Response
Comment 1: Thanks for addressing the comments. No further comments!
Response 1:
Dear Reviewer,
Thank you for your positive feedback and for confirming that the revised manuscript addresses your comments. We sincerely appreciate your time and expertise throughout the review process. Your insights have been invaluable in refining this work.
We are delighted to know that the current version meets your expectations. Please do not hesitate to reach out if any further clarifications are needed.
Once again, thank you for your support and constructive contributions.
Best regards,
Yeqing Wan
On behalf of all authors
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsDear authors,
Thank you for addressing all my remarks and comments. Your manuscript is strongly improved.
One observation I have to make. For my previous comment "Comments 11: I suggest the authors add a Discussion section and Limitations of the study." authors provided a proper analysis of their findings with the latest results. However, this analysis should be moved from the Conclusion section to a different section. The Conclusion section should provide a summary of the results and findings and shouldn't comprise analysis Tables with other studies.
Author Response
Comment 1: One observation I have to make. For my previous comment "Comments 11: I suggest the authors add a Discussion section and Limitations of the study." authors provided a proper analysis of their findings with the latest results. However, this analysis should be moved from the Conclusion section to a different section. The Conclusion section should provide a summary of the results and findings and shouldn't comprise analysis Tables with other studies.
Response 1:
Dear Reviewer,
Thank you for your positive feedback and valuable suggestions. We are pleased to hear that our revisions have significantly improved the manuscript.
In response to your previous comments, we have restructured the Conclusion section to focus solely on summarizing the key findings and future research directions. The comparative analysis tables (Tables 12 and 13) and related discussions have been moved to a newly added Discussion section (Section 6), as suggested. This adjustment ensures that the Conclusion section remains concise while providing a dedicated space for detailed comparisons with existing literature.
We believe these changes align the manuscript more closely with academic standards and enhance the clarity of our contributions. Please let us know if any further refinements are needed.
Thank you once again for your time and constructive input.
Sincerely,
Yeqing Wan
On behalf of all authors
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe revised manuscript is a significant step-up from the original submissions. The authors have provided a suitable rebuttal to some of my initial concerns and the manuscript can proceed for publication.
Author Response
Comment 1: The revised manuscript is a significant step-up from the original submissions. The authors have provided a suitable rebuttal to some of my initial concerns and the manuscript can proceed for publication.
Response 1:
Dear Reviewer,
Thank you for your encouraging feedback and for recognizing the improvements in the revised manuscript. We are deeply grateful for your time and expertise in evaluating our work. Your constructive comments during the initial review phase were instrumental in refining the clarity and scientific rigor of this study.
We are pleased to learn that the revisions adequately address your concerns and that the manuscript is now deemed suitable for publication. Your support throughout this process is greatly appreciated.
Should any further clarifications or adjustments be required, please do not hesitate to reach out.
Once again, thank you for your invaluable contributions to enhancing the quality of this research.
Best regards,
Yeqing Wan
On behalf of all authors
Author Response File: Author Response.pdf