Machine Learning-Based Tourism Demand Prediction Using Tourism Instability Indicators

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear author,

In this manuscript, you explored the impact of a constructed Geopolitical Risk Index on tourism in Saudi Arabia. I appreciate the usage of ML in tourism forecasting; however, I believe that some major flaws are present in this work.

1. The GRI is constructed from tourism dynamics variables and a shock variable (and as far as I can see, only COVID was considered a shock). Later on, this GRI variable is used to predict future tourism dynamics. This would constitute an autoregressive model, labeled as "geopolitical risk". As such, the contributions of this framework are minimal.
2. Tourism volatility can come from many sources, not only geopolitical risk. There are no control variables present.
3. The dataset is extremely small for the methods used and can not be generalised outside of Saudi Arabia.

In the present form, I can't endorse the publication of this article, unless more explanations are given and/or the model is improved.

Author Response

Comment 1

1. The GRI is constructed from tourism dynamics variables and a shock variable (and as far as I can see, only COVID was considered a shock). Later on, this GRI variable is used to predict future tourism dynamics. This would constitute an autoregressive model, labeled as "geopolitical risk". As such, the contributions of this framework are minimal.

Response

I sincerely thank the reviewer for this important observation. I agree that constructing the Geopolitical Risk Index (GRI) from tourism-related dynamics may create the appearance of an autoregressive relationship when later used to explain tourism demand. The reviewer’s comment has helped me improve the conceptual clarity and positioning of the study.

Specifically in the Abstract, I Replace: “constructs a novel Geopolitical Risk Index (GRI) derived from tourism-based variables”

with:

“constructs a tourism-instability proxy index derived from tourism volatility, spending fluctuations, and disruption periods to represent uncertainty conditions affecting tourism demand.

In another update, I revise the Introduction / Contribution Section, and Add a paragraph near the end of the Introduction: It is important to note that the proposed Geopolitical Risk Index (GRI) does not represent a direct geopolitical measurement comparable to established global geopolitical indices. Instead, it functions as a tourism-system instability proxy engineered from internal tourism dynamics, including volatility, disruption periods, and spending fluctuations. The objective is to model how instability signals embedded within tourism behavior influence future tourism demand in data-constrained environments where direct geopolitical indicators are unavailable.

In the other update, I revise the Limitation Section and add the following: Another limitation is that the proposed GRI is partially endogenous to tourism-system behavior because it is constructed from tourism-related volatility indicators. Consequently, the framework may reflect autoregressive tourism instability patterns rather than purely exogenous geopolitical shocks. Future studies should integrate external geopolitical indices, economic uncertainty measures, and security datasets to improve the independence and interpretability of the risk construct.

 

Comment 2

1. Tourism volatility can come from many sources, not only geopolitical risk. There are no control variables present.

Response

I sincerely thank the reviewer for this insightful comment. I fully agree that tourism volatility may originate from multiple factors beyond geopolitical risk, including economic conditions, exchange rates, pandemics, climate variability, seasonality, airline connectivity, policy changes, and destination marketing activities. The reviewer is correct that the current framework does not explicitly include sufficient external control variables. Even though the primary objective of this study was to investigate whether instability-related patterns embedded within tourism-system dynamics could contribute to tourism demand prediction in a data-constrained environment. However, I now clarify in the Methodology Section, specifically in section 4.2 that It is acknowledged that tourism volatility may arise from multiple interacting factors beyond geopolitical conditions, including macroeconomic changes, pandemics, climate variability, exchange-rate fluctuations, transportation accessibility, and destination policies. Due to dataset limitations, the current framework focuses on tourism-derived instability signals rather than a fully controlled causal specification. Therefore, the constructed index should be interpreted as a proxy instability indicator reflecting tourism-system uncertainty

 

Comment 3

1. The dataset is extremely small for the methods used and cannot be generalized outside of Saudi Arabia.

Response

I sincerely thank the reviewer for this important observation. I agree that the dataset size is relatively limited for advanced machine learning applications and that the findings should not be generalized beyond the Saudi Arabian tourism context without caution.

The primary objective of this study was not to develop a universally generalizable forecasting model, but rather to propose and evaluate an exploratory instability-aware tourism analytics framework using the available Saudi Arabia Tourism Dataset (2015–2024). The study was specifically designed as a country-level proof-of-concept under a data-constrained environment associated with Saudi Arabia’s tourism transformation under Vision 2030.

Reviewer 2 Report

Comments and Suggestions for Authors

The only think I can say, it´s I like it.

Author Response

Comment:  The only think I can say, it´s I like it.
Response: Thank you very much

Reviewer 3 Report

Comments and Suggestions for Authors

1.The Introduction should present several clearly defined research questions.

 

2.The theoretical basis and measurement validity of the core variable, the Geopolitical Risk Index (GRI), are insufficient.

 

The most important theoretical and methodological issue in the manuscript concerns the construction of the so-called “Geopolitical Risk Index.” This index is not derived from external geopolitical risk data, but is constructed from internal tourism-related variables. The manuscript explicitly states that the GRI is composed of indicators such as a shock dummy, arrival volatility, and revenue volatility. This approach requires much stronger theoretical justification.

 

Fluctuations in tourist numbers and tourism spending may reflect instability in the tourism market, but they do not necessarily represent geopolitical risk. A decline in tourism demand may result from many factors, such as seasonality, pandemics, economic fluctuations, visa policies, marketing strategies, religious tourism cycles, transportation conditions, price changes, or changes in statistical definitions. Without validation using external geopolitical events or established risk indicators, defining geopolitical risk solely on the basis of fluctuations in tourism variables may lead to conceptual misclassification.

 

I therefore suggest that the authors redefine the theoretical meaning of the GRI and clarify its relationship with the conventional concept of geopolitical risk. If the authors are unable to introduce an external geopolitical risk index, the variable should be renamed more cautiously as a “tourism volatility index” or “tourism instability proxy,” rather than being directly referred to as a “Geopolitical Risk Index.”

 

3.The causal language is too strong and should be revised into predictive or associative language.

 

The manuscript frequently uses terms such as “impact,” “influence,” “confirm,” “validates,” and “plays a significant role,” which may give readers the impression that causal inference has been established. However, based on the current research design, the study is primarily a machine learning-based prediction study and does not employ methods capable of supporting causal identification.

 

For example, in the Results and Discussion sections, the authors interpret the performance of the Stage 2 model as evidence that the GRI has an important influence on tourism demand. However, predictive performance, feature importance, or an increase in R² in a machine learning model can only indicate predictive association among variables. These results do not prove that geopolitical risk has a causal effect on tourism demand.

 

I suggest that the authors substantially weaken causal expressions throughout the manuscript. Terms such as “impact,” “influence,” and “effect” should be replaced with more cautious expressions, such as “association,” “predictive contribution,” “predictive relevance,” or “relationship.” Unless the authors introduce exogenous shocks, event-study methods, Granger causality tests, VAR models, DID designs, or other causal identification approaches, causal conclusions should not be drawn.

 

4.The section numbering throughout the manuscript should be carefully checked.

 

For example, Section 3.1 appears both around Line 258 and Line 336, and Section 4.2 appears both around Line 503 and Line 535. The authors should carefully review and correct the numbering of all sections and subsections.

 

5.The English language requires comprehensive polishing.

 

There are multiple expressions in the manuscript that do not conform to standard academic English, such as “dwells on,” “The fourth layers deal with,” and “The details yearly trends within the datasets reveals.” The authors should thoroughly revise the language throughout the manuscript to improve clarity, grammar, and academic style.

 

 

Comments on the Quality of English Language

The English language should be carefully reviewed.

Author Response

Comment 1

Comments and Suggestions for Authors

1.The Introduction should present several clearly defined research questions.

 Response

I sincerely thank the reviewer for this valuable suggestion. I agree that explicitly stated research questions would improve the clarity, structure, and scientific direction of the paper. The main research question dwell on “How effectively can machine learning models predict the constructed tourism-instability proxy index using tourism dynamics variables” This has been added, along the other research question: “What insights can instability-aware tourism analytics provide for tourism planning and sustainable tourism development under uncertainty?”  

Comment 2

2.The theoretical basis and measurement validity of the core variable, the Geopolitical Risk Index (GRI), are insufficient.

Response

I sincerely thank the reviewer for this important and constructive comment. I agree that the theoretical grounding and measurement validity of the Geopolitical Risk Index (GRI) required clearer justification and stronger conceptual positioning. To address this concern, the paper has introduced theoretical justification, where the proposed GRI has been described not intended to represent a direct geopolitical risk index comparable to established external indices such as Caldara and Iacoviello’s GPR Index. Instead, the variable is now explicitly defined as the theoretical foundation of the proposed tourism-instability proxy is grounded in tourism uncertainty theory and perceived security behavior. Prior tourism studies suggest that instability-sensitive tourism systems often exhibit measurable behavioral fluctuations through changes in arrivals, spending patterns, occupancy variability, and disruption responses. In the absence of direct geopolitical indicators, the present study operationalizes instability indirectly through tourism-system volatility measures, assuming that uncertainty conditions are partially reflected in observable tourism dynamics.

Comment 3

The most important theoretical and methodological issue in the manuscript concerns the construction of the so-called “Geopolitical Risk Index.” This index is not derived from external geopolitical risk data, but is constructed from internal tourism-related variables. The manuscript explicitly states that the GRI is composed of indicators such as a shock dummy, arrival volatility, and revenue volatility. This approach requires much stronger theoretical justification.

 

Response

I sincerely thank the reviewer for this highly valuable and theoretically important comment. I agree that the construction of the proposed Geopolitical Risk Index (GRI) required substantially stronger theoretical clarification and methodological positioning.

The reviewer is correct that the proposed index is not derived from external geopolitical event datasets or established geopolitical risk indicators. Instead, the construct was developed as a tourism-system instability proxy based on internal tourism dynamics, including shock indicators, arrival volatility, and spending volatility. That is exactly the study direction. Specifically, the intention of the study was not to reproduce a conventional geopolitical risk index, but rather to operationalize instability-sensitive tourism behavior under uncertainty conditions in a data-constrained environment. The paper was updated and add a dedicated theoretical justification Inserted within Equations (1) and (2) and also after equation 2

 

Comment 4

Fluctuations in tourist numbers and tourism spending may reflect instability in the tourism market, but they do not necessarily represent geopolitical risk. A decline in tourism demand may result from many factors, such as seasonality, pandemics, economic fluctuations, visa policies, marketing strategies, religious tourism cycles, transportation conditions, price changes, or changes in statistical definitions. Without validation using external geopolitical events or established risk indicators, defining geopolitical risk solely on the basis of fluctuations in tourism variables may lead to conceptual misclassification.

Response

I sincerely thank the reviewer for this highly insightful and theoretically important comment. I fully agree that fluctuations in tourist arrivals and tourism spending do not exclusively represent geopolitical risk and may instead reflect multiple interacting economic, environmental, behavioral, seasonal, policy-related, and operational factors.

 

The reviewer is correct that tourism volatility alone cannot be considered a direct or validated measurement of exogenous geopolitical risk. In its original form, the paper may have overstated the interpretation of the constructed variable. I appreciate this observation and have substantially revised the paper to improve conceptual clarity and methodological transparency. I inserted the justification of the work at near Equation 2, to specifically indicate that It is important to emphasize that fluctuations in tourism arrivals and tourism spending may emerge from multiple interacting factors beyond geopolitical conditions, including seasonal variation, pandemics, economic fluctuations, transportation accessibility, tourism policy changes, pricing effects, and behavioral tourism cycles. Therefore, the proposed construct should not be interpreted as a direct measurement of geopolitical risk, but rather as a tourism-system instability proxy reflecting uncertainty-sensitive tourism dynamics.

Comments 5

I therefore suggest that the authors redefine the theoretical meaning of the GRI and clarify its relationship with the conventional concept of geopolitical risk. If the authors are unable to introduce an external geopolitical risk index, the variable should be renamed more cautiously as a “tourism volatility index” or “tourism instability proxy,” rather than being directly referred to as a “Geopolitical Risk Index.”

Response

I sincerely thank the reviewer for this highly constructive recommendation. I fully agree that the original terminology “Geopolitical Risk Index (GRI)” may overstate the interpretability of the constructed variable, particularly because the index is derived from internal tourism-system dynamics rather than externally validated geopolitical datasets. Clearly following the suggestion of this reviewer and the other reviewer, the paper justifies and directed that the main contribution dwell on formulation of GRI. Hence, the ability to come up with something new, within the context of the real-world dataset is my own side of the story. I also accept all the views/point of views on other direction, which open a clue for further research.

Comments 6

3.The causal language is too strong and should be revised into predictive or associative language.

 Response

I sincerely thank the reviewer for this important observation. I agree that statements in the paper used causal terminology that may overstate the interpretability of the findings, particularly given the exploratory, observational, and proxy-based nature of the study. To address this concern, the paper has been comprehensively revised to replace causal language with more appropriate predictive, associative, and correlational terminology.

Comment 7

The manuscript frequently uses terms such as “impact,” “influence,” “confirm,” “validates,” and “plays a significant role,” which may give readers the impression that causal inference has been established. However, based on the current research design, the study is primarily a machine learning-based prediction study and does not employ methods capable of supporting causal identification.

For example, in the Results and Discussion sections, the authors interpret the performance of the Stage 2 model as evidence that the GRI has an important influence on tourism demand. However, predictive performance, feature importance, or an increase in R² in a machine learning model can only indicate predictive association among variables. These results do not prove that geopolitical risk has a causal effect on tourism demand.

Response

I sincerely thank the reviewer for this highly important methodological clarification. I fully agree that predictive machine learning performance does not establish causal inference and that some interpretations in the original paper may have overstated the implications of the findings.  The reviewer is correct that the current study is fundamentally a machine learning-based predictive and exploratory analytics framework rather than a causal identification study. The confusing statement is address first in the Abstract, where in the Current Abstract:

“The findings confirm that while geopolitical risk is difficult to estimate, it plays a significant role in explaining tourism demand variations.”

Replace with:

“The findings suggest that instability-sensitive tourism indicators may contain predictive information associated with tourism demand variations.”

In the methodology section, I clearly clarify that:

The present study is designed as a predictive machine learning framework rather than a causal inference study. Consequently, model performance metrics, feature importance rankings, and predictive associations should not be interpreted as evidence of causal effects between instability indicators and tourism demand.

Similarly, in Section 4.2 Result:

The Current:

“The higher R² compared to Stage 1 highlights that while geopolitical risk is difficult to predict, it significantly improves the explanation of tourism demand variations.”

Replace with:

“The higher predictive performance observed in Stage 2 suggests that the engineered instability proxy is associated with tourism demand patterns within the predictive modeling framework.”

in the Conclusion Section: Specifically, in the Current:

“This confirms the importance of incorporating instability-related variables into tourism models.”

Replace with:

“The findings suggest that instability-sensitive tourism indicators may provide useful predictive information for tourism demand modeling.”

 

Comments 8

I suggest that the authors substantially weaken causal expressions throughout the manuscript. Terms such as “impact,” “influence,” and “effect” should be replaced with more cautious expressions, such as “association,” “predictive contribution,” “predictive relevance,” or “relationship.” Unless the authors introduce exogenous shocks, event-study methods, Granger causality tests, VAR models, DID designs, or other causal identification approaches, causal conclusions should not be drawn.

 Response

I sincerely thank the reviewer for this highly valuable methodological recommendation. I fully agree that the original paper contained causal expressions that were stronger than what can be supported by the current predictive machine learning framework. Accordingly, in the limitation of the study, I justify by the present study is designed as an exploratory predictive analytics framework rather than a causal inference study. Consequently, machine learning performance metrics, feature importance measures, and predictive relationships should not be interpreted as evidence of causal effects. Future studies employing exogenous shocks, event-study analysis, VAR frameworks, Granger causality testing, or quasi-experimental designs may provide stronger causal interpretation.

Comments 9

4.The section numbering throughout the manuscript should be carefully checked.

 For example, Section 3.1 appears both around Line 258 and Line 336, and Section 4.2 appears both around Line 503 and Line 535. The authors should carefully review and correct the numbering of all sections and subsections.

Response

I sincerely thank the reviewer for identifying this formatting and structural issue. I agree that the section numbering in the paper was inconsistent and may affect readability and navigation. I therefore correct all the sections numbering

Comment 10

The manuscript has been carefully reviewed, and all section and subsection numbering has been corrected. Specifically, the duplicated numbering of Section 3.1 and Section 4.2 has been revised to ensure a logical and sequential structure throughout the paper.

Response

I sincerely thank the reviewer for identifying this formatting and structural issue. I agree that the section numbering in the manuscript was inconsistent and may affect readability and navigation. The paper has been carefully reviewed, and all section and subsection numbering has been corrected. Specifically, the duplicated numbering of Section 3.1 and Section 4.2 has been revised to ensure a logical and sequential structure throughout the paper.

 

Comment 11

5.The English language requires comprehensive polishing.

Response

I sincerely thank the reviewer for this valuable comment. I agree that the paper required significant language improvement to enhance clarity, readability, grammatical consistency, and academic presentation.

Comment 12

There are multiple expressions in the manuscript that do not conform to standard academic English, such as “dwells on,” “The fourth layers deal with,” and “The details yearly trends within the datasets reveals.” The authors should thoroughly revise the language throughout the manuscript to improve clarity, grammar, and academic style.

Response

I sincerely thank the reviewer for this important observation. I agree that several expressions in the original manuscript did not conform to standard academic English and required substantial revision to improve clarity, grammar, readability, and overall academic presentation. Accordingly, I’ve done that.

Reviewer 4 Report

Comments and Suggestions for Authors

1. In Stage 2, you use the Geopolitical Risk Index (GRI)—which is constructed from tourism volatility indicators such as the growth rates of tourist numbers and spending—to predict tourism demand, measured by tourist numbers. Does this approach raise a concern of circular reasoning, whereby the independent variable and the dependent variable share underlying information? Could the improved explanatory power for tourism demand in Stage 2 be partially attributed to this information overlap rather than to the GRI genuinely capturing exogenous geopolitical risk?
2. The GRI is derived entirely from internal tourism system variables (volatility in arrivals, spending volatility, and shock dummies), yet the manuscript provides no external validation demonstrating that this index correlates with actual geopolitical events (e.g., regional conflicts, diplomatic crises). How can you substantiate that this internally derived index effectively proxies the exogenous construct of “geopolitical risk” rather than merely reflecting endogenous supply-and-demand fluctuations within the tourism market?
3. In Stage 1, the GRI prediction achieves an R² of only 0.184 (Extra Trees model), which you attribute to the complexity and indirectly constructed nature of geopolitical risk. However, could this low explanatory power instead indicate that the selected features (lag variables, growth rates, etc.) do not contain sufficient information to predict GRI variations? If the predicted GRI from Stage 1 carries substantial uncertainty, what is the methodological justification for incorporating it into the tourism demand model in Stage 2?
4. In the conceptualization section, you propose a Seasonal-Weather Index (SWI) that combines monthly, quarterly, peak-season, and temperature effects to capture seasonal demand fluctuations. However, the subsequent empirical analysis, feature engineering, and results sections do not describe how the SWI was actually constructed, how its weights were determined, or how it was incorporated into the two-stage modeling framework. Was this conceptual component ultimately integrated into the models? If not, how does its absence affect the completeness of the overall conceptual model?
5. The feature importance analysis in Figure 5(b) highlights “Security Perception Proxy” as an influential predictor, yet the manuscript does not clearly explain how this proxy was derived from the available tourism data, how it differs conceptually and operationally from the GRI, or why both variables are necessary within the same predictive model. Could you elaborate on the construction methodology for this proxy and discuss whether its coexistence with the GRI might introduce redundancy or collinearity issues?
6. You employ a time-based train-test split (training on 2015–2021 and testing on 2022–2024), but the manuscript does not present any robustness checks, such as temporal cross-validation or evaluation across different time windows, to assess the stability of the models over time. Given that geopolitical dynamics and tourism patterns may be subject to structural changes, how confident can one be in the predictive stability of the proposed models across different periods, and is there a risk of overfitting to the specific characteristics of the 2022–2024 test window?

Author Response

Comment 1

Comments and Suggestions for Authors

1. In Stage 2, you use the Geopolitical Risk Index (GRI)—which is constructed from tourism volatility indicators such as the growth rates of tourist numbers and spending—to predict tourism demand, measured by tourist numbers. Does this approach raise a concern of circular reasoning, whereby the independent variable and the dependent variable share underlying information? Could the improved explanatory power for tourism demand in Stage 2 be partially attributed to this information overlap rather than to the GRI genuinely capturing exogenous geopolitical risk?

Response

I sincerely thank the reviewer for this important methodological observation. I acknowledge that the proposed Tourism Instability Proxy (TIP) is constructed using tourism-system indicators such as volatility and growth fluctuations, which are related to tourism demand dynamics. Consequently, some degree of shared informational structure may exist between the predictor variables and the tourism demand target. However, I want to clarify that the proposed framework does not directly reuse raw tourism demand values as predictors in a trivial or duplicated manner. Instead, the instability proxy is derived from transformed temporal dynamics, including volatility measures,

The purpose of Stage 2 is therefore not to establish exogenous causal inference, but rather to evaluate whether instability-sensitive tourism dynamics contain predictive relevance for tourism demand modeling. In time-series analytics and predictive machine learning research, transformed volatility indicators, lag structures, and behavioral dynamics are commonly used to represent latent system instability and uncertainty conditions.

 

 

Comments

1. The GRI is derived entirely from internal tourism system variables (volatility in arrivals, spending volatility, and shock dummies), yet the manuscript provides no external validation demonstrating that this index correlates with actual geopolitical events (e.g., regional conflicts, diplomatic crises). How can you substantiate that this internally derived index effectively proxies the exogenous construct of “geopolitical risk” rather than merely reflecting endogenous supply-and-demand fluctuations within the tourism market?

Response

I sincerely thank the reviewer for this highly important conceptual and methodological observation. I acknowledge that the proposed Tourism Instability Proxy (TIP) is constructed entirely from internal tourism-system variables and does not directly incorporate externally validated geopolitical datasets or independently observed geopolitical event indicators. That is exactly true and to be specific it is from the DATASET that were obtained in real-life not synthetic, the motivation was to extract silent concept from the dataset. I will also like to clarify that the objective of the study was not to claim that the proposed proxy is equivalent to conventional geopolitical risk indices, such as externally constructed geopolitical event-based indices. Instead, the framework was designed to operationalize instability-sensitive tourism dynamics under uncertainty conditions within a data-constrained tourism environment.

 

Comments

1. In Stage 1, the GRI prediction achieves an R² of only 0.184 (Extra Trees model), which you attribute to the complexity and indirectly constructed nature of geopolitical risk. However, could this low explanatory power instead indicate that the selected features (lag variables, growth rates, etc.) do not contain sufficient information to predict GRI variations? If the predicted GRI from Stage 1 carries substantial uncertainty, what is the methodological justification for incorporating it into the tourism demand model in Stage 2?

Response

I sincerely thank the reviewer for this insightful methodological observation. I agree that the relatively low Stage 1 predictive performance (R² = 0.184) indicates that the proposed Tourism Instability Proxy (TIP) is difficult to predict accurately using the selected tourism-system features alone.

However, clarifying this, that the objective of Stage 1 was not to produce a highly deterministic or fully accurate estimation of instability conditions. Instead, Stage 1 was designed as an exploratory feature-engineering and latent instability representation stage intended to capture partial instability-sensitive dynamics embedded within tourism-system behavior.

 

Comments

 

1. In the conceptualization section, you propose a Seasonal-Weather Index (SWI) that combines monthly, quarterly, peak-season, and temperature effects to capture seasonal demand fluctuations. However, the subsequent empirical analysis, feature engineering, and results sections do not describe how the SWI was actually constructed, how its weights were determined, or how it was incorporated into the two-stage modeling framework. Was this conceptual component ultimately integrated into the models? If not, how does its absence affect the completeness of the overall conceptual model?

Response

I sincerely thank the reviewer for this important observation regarding the consistency between the conceptual framework and the empirical implementation. The reviewer is correct that the Seasonal-Weather Index (SWI) introduced in the conceptualization section was not fully operationalized as an explicitly engineered standalone variable within the final two-stage machine learning framework. In the current implementation, certain temporal characteristics associated with seasonality were partially represented through time-related variables, lag structures, and tourism fluctuation patterns. However, the complete SWI formulation, including explicit weighting of monthly, quarterly, peak-season, and temperature effects, was not fully integrated into the empirical modeling pipeline.

 

Comments

 

1. The feature importance analysis in Figure 5(b) highlights “Security Perception Proxy” as an influential predictor, yet the manuscript does not clearly explain how this proxy was derived from the available tourism data, how it differs conceptually and operationally from the GRI, or why both variables are necessary within the same predictive model. Could you elaborate on the construction methodology for this proxy and discuss whether its coexistence with the GRI might introduce redundancy or collinearity issues?

Response

I sincerely thank the reviewer for this highly important methodological observation. I agree that the original paper did not sufficiently explain the conceptual distinction, construction methodology, and operational relationship between the Tourism Instability Proxy (TIP) and the Security Perception Proxy (SPP). I added in the Feature Engineering section, that: the Security Perception Proxy (SPP) was constructed using rolling stability-oriented tourism indicators intended to represent perceived tourism-system confidence and continuity behavior. Unlike the Tourism Instability Proxy (TIP), which emphasizes volatility and disruption-sensitive fluctuations, the SPP emphasizes temporal stability and consistency within tourism activity patterns.

 

 

Comments

 

Response

1. You employ a time-based train-test split (training on 2015–2021 and testing on 2022–2024), but the manuscript does not present any robustness checks, such as temporal cross-validation or evaluation across different time windows, to assess the stability of the models over time. Given that geopolitical dynamics and tourism patterns may be subject to structural changes, how confident can one be in the predictive stability of the proposed models across different periods, and is there a risk of overfitting to the specific characteristics of the 2022–2024 test window?

Response

I sincerely thank the reviewer for this important methodological observation. I agree that relying only on a single chronological train-test split may limit confidence in the temporal stability of the proposed models. The original design used a time-based split, with 2015–2021 as the training period and 2022–2024 as the testing period, to avoid data leakage and preserve the temporal order of the tourism dataset. However, I acknowledge that the chronological train-test split was adopted to preserve temporal ordering and reduce data leakage. However, the present study does not perform rolling-window or walk-forward temporal cross-validation. Therefore, the reported results should be interpreted as predictive performance under a single chronological forecasting setting rather than as evidence of stable model performance across all possible periods.

Reviewer 5 Report

Comments and Suggestions for Authors

This paper constructs a Geopolitical Risk Index (GRI) from tourism data and adopts a two‑stage machine learning framework to predict tourism demand in Saudi Arabia. The research perspective is innovative with a rigorous design and reliable data. Overall, this is a medium‑to‑upper level empirical paper suitable for publication after minor revisions.

(1) Supplement visualization analysis of model residuals and prediction errors. The empirical section only reports fitting indicators and lacks detailed error analysis. Please supplement residual distribution plots and annual prediction error line charts for all models to visually demonstrate the fitting performance.

(2) Verify the consistency between the GRI and real geopolitical events. The paper does not test whether the constructed GRI can represent real geopolitical risks, which weakens the credibility of the core conclusions. Please match the GRI time series with real events including regional conflicts, the COVID‑19 pandemic, and policy adjustments in Saudi Arabia, and verify the index validity using 2–3 typical events.

(3) Refine policy recommendations for Saudi Arabia. The existing policy recommendations are general and not closely integrated with the research findings. Based on the high‑volatility periods of the GRI, please propose targeted measures including risk early warning mechanisms, visitor flow regulation in key provinces such as Makkah, and off‑season marketing optimization.

Author Response

Comment 1

Comments and Suggestions for Authors

This paper constructs a Geopolitical Risk Index (GRI) from tourism data and adopts a two‑stage machine learning framework to predict tourism demand in Saudi Arabia. The research perspective is innovative with a rigorous design and reliable data. Overall, this is a medium‑to‑upper level empirical paper suitable for publication after minor revisions.

Response

I sincerely thank the reviewer for the positive and encouraging evaluation of the paper. I greatly appreciate the reviewer’s recognition of the novelty of the research perspective, the structured two-stage machine learning framework, and the overall empirical contribution of the study. The reviewer’s detailed comments and methodological observations have been extremely valuable

 

Comment 2

(1) Supplement visualization analysis of model residuals and prediction errors. The empirical section only reports fitting indicators and lacks detailed error analysis. Please supplement residual distribution plots and annual prediction error line charts for all models to visually demonstrate the fitting performance.

Response

I sincerely thank the reviewer for this valuable suggestion. I agree that aggregate performance metrics alone may not fully capture model behavior and prediction quality. The suggested plots from the reviewer are captured in Figure 4, which confirms that volatility and lag effects are key determinants in modeling geopolitical risk proxies.

Comment 3

(2) Verify the consistency between the GRI and real geopolitical events. The paper does not test whether the constructed GRI can represent real geopolitical risks, which weakens the credibility of the core conclusions. Please match the GRI time series with real events including regional conflicts, the COVID‑19 pandemic, and policy adjustments in Saudi Arabia, and verify the index validity using 2–3 typical events.

Response

 

Comment 4

(3) Refine policy recommendations for Saudi Arabia. The existing policy recommendations are general and not closely integrated with the research findings. Based on the high‑volatility periods of the GRI, please propose targeted measures including risk early warning mechanisms, visitor flow regulation in key provinces such as Makkah, and off‑season marketing optimization.

Response

I sincerely thank the reviewer for this important observation. We acknowledge that the proposed Tourism Instability Proxy (TIP) was not directly validated against externally established geopolitical event datasets or formal geopolitical risk indices. However, I clarify that the objective of the present study was not to construct a conventional geopolitical risk index comparable to externally validated geopolitical-event measures. Instead, the framework was intentionally designed as a tourism-system instability representation fo

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,

I will accept this revision after changing the titles and minimising the importance of "geopolitical risk", since it wasn't actually modelled.

Thank you for your time.

Author Response

Comment 1

I will accept this revision after changing the titles and minimising the importance of "geopolitical risk", since it wasn't actually modelled.

Response

Thank you very much for the suggestion, the title has been changed based on your views as:

“Machine Learning-Based Tourism Demand Prediction Using Tourism Instability Indicators” to recalibrate the importance of "geopolitical risk", references to “Geopolitical Risk Index (GRI)” as a tourism-instability indicator.

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript has undergone substantial revisions; however, the quality of the English still requires further improvement. The authors should carefully proofread the entire manuscript again. For example, in the Abstract, line 12, the sentence “to represent uncertainty conditions affecting tourism demand. including volatility…” contains an inappropriate full stop in the middle of the sentence, which results in a sentence fragment. In the Introduction, line 35, the word “sec-tor” should also be corrected. In addition, in the response letter, the authors are advised to indicate the corresponding page and line numbers for each revision so that the changes can be more easily identified and checked by the reviewers.

Comments on the Quality of English Language

The English quality of the manuscript still needs to be improved, and the authors should carefully proofread the entire manuscript again. For example, in the Abstract, line 12, the sentence “to represent uncertainty conditions affecting tourism demand. including volatility…” contains an inappropriate full stop in the middle of the sentence, resulting in a sentence fragment. In the Introduction, line 35, the word “sec-tor” should also be corrected. These examples indicate that the manuscript still contains language and formatting issues, and a thorough English language revision is recommended before further consideration.

Author Response

Comment 1

Comments and Suggestions for Authors

The manuscript has undergone substantial revisions; however, the quality of the English still requires further improvement. The authors should carefully proofread the entire manuscript again. For example, in the Abstract, line 12, the sentence “to represent uncertainty conditions affecting tourism demand. including volatility…” contains an inappropriate full stop in the middle of the sentence, which results in a sentence fragment. In the Introduction, line 35, the word “sec-tor” should also be corrected. In addition, in the response letter, the authors are advised to indicate the corresponding page and line numbers for each revision so that the changes can be more easily identified and checked by the reviewers.

Response

Thank you very much for the suggestion.The corrections have been done. For the abstract in Line 12 has been change and the changes is indicated with green highlight.  Similarly, in the introduction, the green line of this particular correction has been included.

Comment 2

Comments on the Quality of English Language

The English quality of the manuscript still needs to be improved, and the authors should carefully proofread the entire manuscript again. For example, in the Abstract, line 12, the sentence “to represent uncertainty conditions affecting tourism demand. including volatility…” contains an inappropriate full stop in the middle of the sentence, resulting in a sentence fragment. In the Introduction, line 35, the word “sec-tor” should also be corrected. These examples indicate that the manuscript still contains language and formatting issues, and a thorough English language revision is recommended before further consideration.

Response

Thank you very much for the suggestion. A general proofreading has been done

Reviewer 4 Report

Comments and Suggestions for Authors

The authors have completed the revisions and have addressed the comments appropriately. I recommend acceptance.

Author Response

Comment 1

Comments and Suggestions for Authors

The authors have completed the revisions and have addressed the comments appropriately. I recommend acceptance.

Response

Thank you very much