How the Pandemic Changes the Factors Influencing Aircraft Utilization: The Case of Korea

Abstract

We investigate how the factors influencing aircraft utilization have changed during and post-Pandemic depending on the business model before. We classify the Pandemic into three periods (pre-, during and post- Pandemic) and the business models into three types (Total, FSC and LCC). For each group, we analyze the importance of factors using the SHAP and Random Forest models. Through group-difference tests on factor importance, we examine whether there are significant differences across the three periods and business models. According to the findings of the ANOVA (Analysis of Variance) and the Kruskal–Wallis assay, the importance of factors influencing aircraft utilization has changed across all business models over the three periods. Pre-Pandemic, a coincident index and a consumer price index were the principal factors. However, the exchange rate (KRW/EUR) gained significant importance during the Pandemic. This suggests that the Pandemic’s impact on the aviation industry was not limited to reduced demand but was also associated with changes in the importance of exchange rates and key business indicators for airline operations. Pre-Pandemic, there were significant differences among the business model groups. However, no meaningful differences were observed during and post-Pandemic. In other words, it seems that the leading indexes were closely interconnected pre-Pandemic, whereas lagging indexes and exchange rate became closely interconnected afterward. A group-difference test confirmed that no differences were observed among the business models, but differences were evident when considering the groups in their entirety. We presented the implications for changes in airline decision-making to understand changes in the aviation industry caused by the Pandemic, by identifying how the factors influencing aircraft utilization were altered.

1. Introduction

We verify a hypothesis that the Pandemic changed the importance of factors influencing aircraft utilization by airlines and a hypothesis that the Pandemic also changed the importance of factors influencing the business models. This study explores how socio-economic factors affecting aircraft utilization shifted during the COVID-19 Pandemic, across airline business models. Aircraft utilization is a critical factor in generating airline revenue (Homsombat, W., Lei, Z., & Fu, X. [1]), and understanding whether and how the importance of influencing factors changes based on the combination of the business model and the pre-/post-Pandemic periods is crucial for comprehending supply related decision-making by airlines. Some studies have focused on optimizing aircraft operation according to the fleet composition (Petersen, Jon D. et al. [2]). However, there has been no research on aircraft operation in relation to changes in external environment influencing the decision-making of airlines. In particular, understanding how the factors influencing aircraft utilization have changed due to external environmental changes since the Pandemic would provide significant implications for the air transport industry.

As societal and economic uncertainty increases, airlines experience greater difficulty in establishing plans for aircraft utilization by route and realize the importance of comparative advantage strategies. However, time series analysis and similar methods, which assume the balance of supply and demand, are limited in establishing the plans and strategies for aircraft utilization in situations where uncertainty spreads rapidly and extensively. Earlier research (Hanlon, P. [3]; Madhavan, M et al. [4]) focused on factors influencing aircraft utilization. Although previous studies have offered valuable insights into airline recovery and demand disruptions in the post-COVID-19 context, the specific structural and operational changes brought about by the Pandemic remain underexplored.

We analyzed factors grouping variables that influence the strategy for aircraft utilization by classifying the business models into pre- and post-Pandemic. It is known that certain variables have a significant influence on air travel demand (Abrahams, M. [5]; Brons, M. et al. [6]). Aircraft utilization is influenced by various factors in a complex way, as it depends on air travel demand, which is sensitive to specific variables, and a supplier’s decision-making regarding the external environment. It is necessary to review the traditional understanding of aircraft utilization, given the significant impact of the Pandemic on the aviation industry. We classified and reviewed the factors influencing aircraft utilization based on the Pandemic period and business models, using the SHAP methodology and the Random Forest model for exploratory data analysis.

It is well known that short-term economic indicators, exchange rates, and similar factors influence both air travel demand factors (e.g., passengers, tariffs, convenience-related factors, etc.) and supply factors (e.g., traffic rights, infrastructure, slot-related factors, etc.) (Yang, T. et al. [7]; Wu, C. Y.et al. [8]). It was anticipated that such external factors would influence aircraft utilization even during the Pandemic. However, it was assumed that each indicator would affect aircraft utilization in levels and methods different from those pre-Pandemic. To address the heightened uncertainty and nonlinear dynamics introduced by the COVID-19 Pandemic, machine learning methods such as Random Forest and SHAP provide robust tools for modeling complex relationships and identifying key variables, complementing traditional econometric approaches. This study performed exploratory data analysis using SHAP analysis (Pace, R. K., & Barry, R. [9]; Komatsu, M. et al. [10]) and Random Forest analysis (Breiman, L. [11]), rather than confirmatory data analysis of aircraft utilization, causality, or correlation of independent variables, to identify external influencing factors and verify changes.

In Section 2, we reviewed preceding research on aircraft utilization strategy, SHAP analysis, Random Forest analysis, group test analysis, and related topics. In Section 3, we explained the methodologies for analyzing influencing factors applied in the present research. In Section 4, we examined our hypothesis regarding business models in terms of periods, such as pre-, during, and post-Pandemic. Lastly, in Section 5, we presented the discussions and conclusion.

2. Literature Review

2.1. Prospects for the Air Transport Industry and Tourism Industry After COVID-19

The air transport industry and tourism industry have suffered greatly due to the decrease in air passengers caused by COVID-19, which emerged in 2019. Many studies have analyzed and prospected such industries since the Pandemic.

Numerous studies have evaluated the effects that the Pandemic had on the air transport and tourism industries and analyzed recovery strategies, as follows: Sun et al. [12] analyzed the impact the Pandemic had on the aviation industry based on a survey conducted among ATRS participants and proposed recovery measures. And Gagan Deep Sharma et al. [13] presented a recovery framework focused on four major elements (governmental response, technical innovation, local membership, reliability) for the recovery of the tourism industry post-Pandemic. Khaled Halteh et al. [14] analyzed the effects of COVID-19 on general aviation (GA), through a literature review and interviews of industry participants. Kaitano Dube et al. [15] analyzed the overall effects of the Pandemic on the aviation industry and explored a recovery path through data and case analysis.

Such recovery measures were supplemented with data analysis and quantitative approaches. Sun et al. [16] presented data-driven recovery strategies for the aviation industry post-Pandemic and identified major recovery factors. Khaled Halteh et al. [14] assessed the financial effects of COVID-19 on the aviation industry, using machine learning techniques, such as Random Forests. Demola Obembe et al. [17] explored the effects of COVID-19 with an analysis of news and social data from the early stages of the Pandemic and highlighted the role of communication during crises.

The following studies have also initiated discussions on the structural changes and sustainability of the aviation industry caused by the Pandemic. In this study, the term ‘structural changes’ refers to significant transformations in the aviation industry’s regulatory frameworks, operational strategies, and market structures, including regulatory shifts and strategic adjustments in response to the COVID-19 Pandemic. Gokhan Tanriverdi et al. [18] analyzed the strategic response of the aviation industry post-Pandemic, examining the keywords and correlation among research themes from the Journal of Air Transportation Management (JATM). Johannes Michelmann et al. [19] assessed structural changes and sustainability challenges of the aviation industry caused by the Pandemic and presented recovery scenarios. Also, Sun et al. [20] classified the effects of COVID-19 on the aviation industry into early impact, deadlock and recovery phases and investigated their changes by phase.

2.2. Prospects and Analysis of Airlines’ Strategies

Airlines were required to amend their operational strategies based on the prospects of the air transport and tourism industries. Various insights were provided for airlines on topics such as the input of aircrafts; scheduling; and simulation-based efficiency optimization.

The following studies have analyzed the effects of the Pandemic and proposed strategies and recovery scenarios. Wang, Wu, Fu & Wang [21] analyzed the Airline-within-Airline strategy and the Pandemic’s impact on Indonesia’s domestic aviation market, conducting Probit model and multinomial logistic regression analysis. Gualini et al. [22] analyzed various strategies adopted by U.S. airlines during the Pandemic. Arrigo et al. [23] compared and analyzed the strategic responses of European airlines to market changes, including the Pandemic, using various case studies. Xuan et al. [24] analyzed the Pandemic’s effects on the aviation industry, forecasting profits based on the VAR model.

The following studies sought to improve the airlines’ optimization strategies and operational efficiency through various methodologies based on the effects of the Pandemic. Štimac et al. [25] simulated and analyzed the effects of airlines’ business models on the efficiency of airport operations and capacity optimization using the AMSS. Enki et al. [26] developed a capacity planning methodology to maximize aircraft input and network contribution and proposed a method to resolve aircraft input issues. Also, Eufrasio et al. [27] assessed an effect of the schedule padding strategy using the high dimensional sparse (HDS) regression model with a focus on airlines in Brazil. And Atay, Seckiner & Eroglu [28] examined how to maximize the aircraft operational strategies of airlines using the response surface methodology and analyzing boarding rates at a break-even point.

The following studies analyzed air traffic scenarios post-Pandemic and presented network optimization strategies for recovery. Kiziloğlu & Sakalli. [29] applied a meta-heuristic algorithm based on the Monte Carlo simulation and developed a model integrating the flight schedule, aircraft input and aircraft route optimization in a code share situation. Gelhausen et al. [30] analyzed the global air traffic scenarios post-Pandemic until 2040 and proposed strategies for deregulating airport capacity based on the best- and worst-case recovery scenarios.

2.3. Methodologies for Analyzing Grouping Factors

As the industrial structure evolved, the strategies of airlines changed, and the factors influencing air travel demand changed as well. Accordingly, many studies have grouped and explored these factors.

SHAP analysis was applied to understand the relationships and derive the main factors. Ki-Han Song et al. [31] identified short-term air travel demand using social and economic variables and the number of confirmed cases of COVID 19 through SHAP analysis. Verma et al. [32] analyzed demographic, social and economic factors using travel data, multivariate statistics, and SHAP analysis. Li et al. [33] identified the main factors influencing intercity travel mode selection using XGBoost and SHAP.

Additionally, the following studies analyzed the demand factors using machine learning-based methodologies and developed demand prediction models. Jian-Wu Bi et al. [34] developed demand prediction models by converting time series data into images and analyzing them with a deep learning model. And Chanittha Chansuk et al. [35] conducted exploratory and confirmatory data analyses, amending the Theory of Planned Behavior (TPB) using Thailand’s domestic tourist survey data to analyze factors affecting tourist behavior changes caused by COVID-19. Li et al. [36] collected factors determining tourist satisfaction using data mining and analyzed these factors using Random Forest and K-Mode Clustering. Mahmut Bakir et al. [37] derived attributes of airport service that influence airport service satisfaction by reviewing the passengers’ response. Gao et al. [38] identified main factors influencing multimodal travel selection using a machine learning-based ensemble learning model. Wu et al. [39] analyzed patterns and factors of abnormal passenger behavior in subway systems using XGBoost and DELogit-based models. Finally, Deng et al. [40] examined differences in main factors influencing route selection during rush hour through XGBoost and GAM.

2.4. Comparative Test Among Groups

The following studies have been continuously conducted to verify whether the average or rates of specific variables differed among different groups, evaluate group effectiveness, and support decision-making based on these differences through a group-difference test.

Dao et al. [41] used the Wilcoxon rank sum test to monitor the state of wind turbines, set a zero-defect state to be a null hypothesis, and test the null hypothesis. Thus, the state of the wind turbines could be determined through the test. Natarajan et al. [42] analyzed differences in education level, health condition and income level according to insurance subscription using MEPS data from the United States. Bridge P.D. [43] statistically compared the t-test and the Wilcoxon rank sum test to analyze the performance of group-difference tests and evaluation indices.

3. Methodology

First, this research conducted a basic analysis of how the domestic route utilization rate has changed since the Pandemic. The analysis used the explanatory variable as an independent variable and the number of seats supplied domestically as the dependent variable to reliably explore the factors influencing changes in aircraft utilization rate by route. The number of seats supplied refers to the total number of seats airlines or flights can offer during a specific period. It can also be defined as the number of seats available for sale on a specific route, determined by the operational schedule of airlines. The number of seats supplied is an important index for the operation of airlines and seems to be essential for balancing supply and demand. Next, this research intended to derive the importance of factors for the SHapley Additive exPlanations (SHAP) and the Random Forest analyses, which are methodologies for grouping factor analysis. Finally, the research sought to compare the influences on the international and domestic routes pre- and post-Pandemic. The research landscape, as illustrated in Figure 1, is presented as follows.

In terms of the dataset, the number of seats supplied was derived from the Tower-Log data of Korea, and a ratio (%) of the number of seats supplied for international routes was calculated based on each plane number by refining the number of seats supplied by date from 2018 to 2023. The following indexes were set as the socio-economic variables: exchange rate (KRW/USD, KRW/RMB, KRW/100 JPY, KRW/EUR); stock price indexes (KOSPI and KOSDAQ indexes); and business indexes (leading index, coincident index, lagging index).

SHAP is a methodology that analyzes the importance and influence of each factor in a group, based on the Shapley Value in game theory. The Shapley Value calculates the fair contribution of each participant and is defined as a value obtained by adding up the contribution of each characteristic in all possible combinations (Van den Broeck, G. et al. [44]).

The present research aimed to derive the importance of factors influencing aircraft utilization demand by air route with the SHAP methodology. (Lundberg, S. M., & Lee, S. L. [45]) The LightGBM ensemble model (Ke, G. et al. [46]), which forms the basis of the SHAP methodology, used the number of seats supplied, relating to aircraft utilization by air route, as the input variable. It was designed to be a research model that evaluates performance by incorporating stock price, exchange rate and other socio-economic variables to be learning data. Figure 2 illustrates an example of how the SHAP methodology can be applied.

$${\varphi }_i=\sum _{S\subseteq F\backslash i}{\displaystyle \frac{\left|S\right|!\left(\left|F\right|-\left|S\right|-1\right)!}{\left|F\right|!}}\left[f_{S\cup i}\left(x_{S\cup i}\right)-f_S\left(x_S\right)\right]$$

${\varphi }_i$ = Shapley value for i-th data;

$F$ = Complete set;

$S$ = All remaining subsets of the complete set excluding the i-th data;

$f_{S\cup i}\left(x_{S\cup i}\right)$ = Total contribution including the $i$-th data;

$f_S\left(x_S\right)$ = Contribution of the remaining subsets excluding the i-th data.

Figure 2. Example of how to apply the SHAP methodology.

Figure 2. Example of how to apply the SHAP methodology.

The Random Forest is an ensemble learning-based model composed of multiple decision trees used for classifying and regressing specific events. It is primarily applied to solve classification and regression problems (Rigatti, S. J. [47]). Each tree is built using random samples of data. One key advantage of the Random Forest is the Permutation Importance function, which measures variable importance by observing changes in model performance when variable values are randomly shuffled. If performance decreases significantly, the variable is considered important (Huang, N., Lu, G., & Xu, D. [48]). The predictive value of the Random Forest model is expressed in the following formula (Breiman, L. [11]). Figure 3 presents a representative structure of the Random Forest regression model.

$$\widehat{y}={\displaystyle \frac{1}{N}}\sum _{i=1}^N{h}_i\left(x\right)$$

$\widehat{y}$ = Final predictive value;

$N$ = Number of trees;

$h_i\left(x\right)$ = Predictive value of i-th decision tree;

$x$ = Value of input characteristics vector.

The formula for variable importance in the Random Forest model is expressed in the following formula (Breiman, L. [11]).

$$I_j={\displaystyle \frac{1}{N}}\sum _{t=1}^{N_t}({Impurity}_{before}-{Impurity}_{after})$$

I_j = Importance of variable j;

N_T = The number of all trees;

Impurity_before = Level of impurity before node division (entropy);

Impurity_after = Level of impurity after node division.

Figure 3. A representative arrangement of Random Forest regression structure (Liaw, A. [49]).

Figure 3. A representative arrangement of Random Forest regression structure (Liaw, A. [49]).

We set a research hypothesis based on the importance derived from SHAP and Random Forest analyses and conducted ANOVA and the Kruskal–Wallis assay, which are methods for group-difference tests among three or more independent groups. We defined the research hypothesis as follows:

A null hypothesis for the ANOVA test is defined to be that an average of all groups is identical, i.e., “there is no difference among groups” as follows:

$$\begin{gathered} \mathrm{H}0=\mathrm{\mu }1=\mathrm{\mu }2=\mathrm{\mu }3=\mathrm{\dots }=\mathrm{\mu }\mathrm{k} \\ k=\mathrm{The} \mathrm{number} \mathrm{of} \mathrm{groups} \end{gathered}$$

The null hypothesis for the Kruskal–Wallis assay (Kruskal, W. H., & Wallis, W. A. [50]) relates to the effect that the distribution of all groups is identical, i.e., there is no difference among groups as follows:

$$\mathrm{H}0=\mathrm{The} \mathrm{distribution} \mathrm{of} \mathrm{all} \mathrm{groups} \mathrm{is} \mathrm{identical}$$

ANOVA analyzes differences in average groups among groups to verify whether statistically meaningful differences exist. It is used to assess the effect of a single independent variable on dependent variables (Norton, B. J., & Strube, M. J. [51]). ANOVA derives the significance probability based on F-statistics, enabling hypothesis testing. ANOVA derives the significance probability based on the F-statistics and could test a hypothesis on the significance probability.

$$\begin{gathered} F = {\displaystyle \frac{(\mathrm{S}\mathrm{u}\mathrm{m} \mathrm{o}\mathrm{f} \mathrm{s}\mathrm{q}\mathrm{u}\mathrm{a}\mathrm{r}\mathrm{e}\mathrm{s} \mathrm{a}\mathrm{m}\mathrm{o}\mathrm{n}\mathrm{g} \mathrm{g}\mathrm{r}\mathrm{o}\mathrm{u}\mathrm{p}\mathrm{s}(SSB\left)\right)/(k-1)}{(\mathrm{S}\mathrm{u}\mathrm{m} \mathrm{o}\mathrm{f} \mathrm{s}\mathrm{q}\mathrm{u}\mathrm{a}\mathrm{r}\mathrm{e}\mathrm{s} \mathrm{w}\mathrm{i}\mathrm{t}\mathrm{h}\mathrm{i}\mathrm{n} \mathrm{g}\mathrm{r}\mathrm{o}\mathrm{u}\mathrm{p}\mathrm{s}(SSW)/(N-k)}} \\ N=\mathrm{The} \mathrm{total} \mathrm{number} \mathrm{of} \mathrm{data} \\ k=\mathrm{The} \mathrm{number} \mathrm{of} \mathrm{groups} \end{gathered}$$

The Kruskal–Wallis assay (Kruskal, W. H., & Wallis, W. A. [50]) is a non-parametric statistical test used to compare three or more independent groups. This test is useful when data do not follow a normal distribution and is used assess whether the median values of each group are identical. The test statistics are calculated as follows:

$$\begin{gathered} H = {\displaystyle \frac{12}{N\left(N+1\right)}}\sum _{i=1}^k{\displaystyle \frac{R_i^2}{n_i}} - 3\left(N+1\right) \\ N=\mathrm{The} \mathrm{number} \mathrm{of} \mathrm{all} \mathrm{samples} \\ k=\mathrm{The} \mathrm{number} \mathrm{of} \mathrm{groups} \\ {R}_i=\mathrm{Sum}\mathrm{of}\mathrm{ranking}\mathrm{of} i-\mathrm{th} \mathrm{group} \\ {n}_i=\mathrm{The} \mathrm{number} \mathrm{of} \mathrm{samples} \mathrm{in} i-\mathrm{th} \mathrm{group} \end{gathered}$$

In the realm of the SHAP model, the analysis leverages Shapley Values, which aggregate the contributions of each feature across all possible combinations. This methodology allows for a nuanced evaluation of the importance of individual factors by clarifying their influence on predicted outcomes and quantifying the extent of changes in predictions based on variations in input variables (Van den Broeck. et al. [44]). The Random Forest model is designed to improve the consistency of classification and prediction tasks. It adeptly captures the inherent characteristics of the data, producing robust classifications and models that are instrumental in decision-making processes (Breiman., L. [11]). This capability facilitates the identification of key contributing factors that significantly impact predictions.

For each model, the time was classified as follows: two years from 2018 to 2019, which are pre-Pandemic; two years from 2020 to 2021, which are during the Pandemic; and two years from 2022 to 2023, which are post-Pandemic. For each time period, the daily ratio of seats supplied for international routes of the Total, FSC and LCC was input on a daily data basis. Changes in the SHAP-derived feature importance and the findings from the Random Forest model’s importance analysis were explored, followed by a comprehensive analysis to test for differences among groups.

4. Results

4.1. Findings of Basic Analysis

According to the basic analysis of time series data, the ratios of seats supplied for international routes were approximately 60% and 80% for the LCC and FSC, respectively, pre-Pandemic. During the Pandemic, the ratios declined to approximately 0–10% for the LCC and 40–60% for the FSC. Post-Pandemic, both the FSC and LCC exhibited a recovery trend, and by the end of 2023, their ratios were similar to those observed pre-Pandemic. Figure 4 shows daily seat supply ratios for international routes (total, FSC, and LCC) from 1 January 2018 to 31 December 2023.

4.2. SHAP Analysis

4.2.1. Total (FSC + LCC)

According to the findings of the SHAP analysis of the ratio of the number of seats supplied for international routes, the coincident index, consumer price index and KOSPI index were ranked highly in such order pre-Pandemic. And the exchange rate (KRW/EUR), KOSDAQ index and exchange rate (KRW/100 JYP) were ranked highly in such order during the Pandemic. Also, the lagging index, consumer price index and exchange rate (KRW/USD) were important in such order post-Pandemic.

Overall, SHAP analyses show that the exchange rate (KRW/EUR) gained prominence during the Pandemic, and the lagging index became highly significant post-Pandemic across business models.

Table 1. Findings of comparative analysis of grouping factors (SHAP, Total).

Classification	Pre-Pandemic		During the Pandemic		Post-Pandemic
	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank
KOSDAQ_ Index	4.33%	9	29.98%	2	0.53%	9
KOSPI_ Index	14.28%	3	4.93%	5	3.93%	5
Coincident_Composite_Index	23.99%	1	0.32%	8	0.25%	10
Leading_ Composite_Index	4.70%	8	0.09%	9	51.09%	1
Price_ Index	16.90%	2	2.90%	7	26.65%	2
W/ JPY100	7.78%	6	14.35%	3	2.25%	6
W/ USD	8.20%	5	6.25%	4	7.25%	3
W/ CNY	11.61%	4	3.67%	6	1.68%	7
W/ EUR	7.56%	7	37.50%	1	1.37%	8
Lagging_ Composite_Index	0.64%	10	0.00%	10	5.01%	4

presents the findings of the comparative analysis of grouping factors (SHAP, Total).

4.2.2. FSC

According to the findings of the SHAP analysis of the ratio of the number of seats supplied for the FSC, the leading index, exchange rate (KRW/100 JPY) and exchange rate (KRW/RMB) index were ranked highly in such order pre-Pandemic. The exchange rate (KRW/EUR), KOSDAQ index and consumer price index were ranked highly in such order during the Pandemic. Also, the importance was identical in such order even post-Pandemic.

Notably, it was verified that for the FSC, the exchange rate (KRW/EUR), consumer price index, and KOSDAQ index were highly ranked in importance pre-Pandemic. However, the leading index, which had been highly important pre-Pandemic, experienced a rapid decline in importance afterward.

Table 2. Findings of comparative analysis of grouping factors (SHAP, FSC).

Classification	Pre-Pandemic		During the Pandemic		Post-Pandemic
	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank
KOSDAQ_Index	3.98%	8	19.59%	2	19.25%	2
KOSPI_ Index	3.32%	9	6.68%	5	6.57%	5
Coincident_Composite_Index	7.61%	6	2.13%	8	2.09%	8
Leading_Composite_Index	22.74%	1	0.14%	9	0.14%	9
Price_ Index	6.70%	7	19.52%	3	19.19%	3
W/ JPY100	15.98%	2	6.28%	6	6.18%	6
W/ USD	11.26%	5	6.93%	4	8.52%	4
W/ CNY	14.98%	3	4.95%	7	4.86%	7
W/ EUR	11.62%	4	33.77%	1	33.19%	1
Lagging_Composite_Index	1.82%	10	0.01%	10	0.01%	10

presents the findings of the comparative analysis of grouping factors (SHAP, FSC).

4.2.3. LCC

According to the findings of the SHAP analysis of the ratio of the number of seats supplied for LCC, the KOSPI index, consumer price index and exchange rate (KRW/RMB) index were ranked highly in that order pre-Pandemic. The KOSPI index, exchange rate (KRW/USD) and consumer price index were ranked highly in such order during the Pandemic. Also, the exchange rate (KRW/EUR), KOSDAQ index and consumer price index were important in such order post-Pandemic.

Notably, the KOSDAQ index, which was ranked low for LCC pre- and during the Pandemic, rose to a high-ranking post-Pandemic. In contrast, the KOSPI index, which had been among the highest-ranked factors pre- and during the Pandemic, dropped significantly in importance afterward. In addition, the consumer price index consistently maintained a high ranking both pre- and post-Pandemic.

Table 3. Findings of comparative analysis of grouping factors (SHAP, LCC).

Classification	Pre-Pandemic		During the Pandemic		Post-Pandemic
	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank
KOSDAQ_Index	1.03%	8	0.80%	8	19.25%	2
KOSPI_Index	40.55%	1	31.43%	1	6.57%	5
Coincident_Composite_Index	0.27%	9	0.21%	9	2.09%	8
Leading_Composite_Index	10.55%	4	8.17%	5	0.14%	9
Price_Index	16.61%	2	12.88%	3	19.19%	3
W/JPY100	6.06%	7	4.69%	7	6.18%	6
W/USD	6.67%	6	27.66%	2	8.52%	4
W/CNY	11.48%	3	8.90%	4	4.86%	7
W/EUR	6.78%	5	5.26%	6	33.19%	1
Lagging_Composite_Index	0.00%	10	0.00%	10	0.01%	10

presents the findings of the comparative analysis of grouping factors (SHAP, LCC).

4.3. Random Forest Analysis

4.3.1. Total (FSC + LCC)

According to the findings of the Random Forest analysis of the ratio of the number of seats supplied for international routes for the Total, the consumer price index, exchange rate (KRW/RMB) and lagging index were ranked highly in such order pre-Pandemic. And the exchange rate (KRW/EUR), KOSDAQ index and exchange rate (KRW/100 JYP) were ranked highly in such order during the Pandemic. Also, the lagging index, leading index and coincident index were important in such order post-Pandemic.

According to the findings of the Random Forest analysis, the leading index was ranked low pre-and during the Pandemic but became highly important post-Pandemic. Consistent with SHAP, Random Forest analyses also highlighted the exchange rate (KRW/EUR) and lagging index as key factors during and post-Pandemic, respectively.

Table 4. Findings of comparative analysis of grouping factors (Random Forest, Total).

Classification	Pre-Pandemic		During the Pandemic		Post-Pandemic
	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank
KOSDAQ_Index	5.55%	8	6.32%	2	0.16%	10
KOSPI_Index	5.27%	9	2.58%	5	0.36%	6
Coincident_Composite_Index	10.42%	5	1.07%	7	14.39%	3
Leading_Composite_Index	2.98%	10	0.58%	10	18.64%	2
Price_Index	17.13%	1	1.20%	6	14.18%	4
W/JPY100	11.77%	4	5.47%	3	0.25%	9
W/USD	7.93%	6	3.10%	4	6.02%	5
W/CNY	16.73%	2	0.86%	8	0.36%	6
W/EUR	7.69%	7	78.03%	1	0.30%	8
Lagging_Composite_Index	14.53%	3	0.79%	9	45.34%	1

presents the findings of the comparative analysis of grouping factors (Random Forest, Total).

4.3.2. FSC

According to the findings of the Random Forest analysis of the ratio of the number of seats supplied for international routes for FSC, the lagging index, exchange rate (KRW/100 JPY) and lagging index were ranked highly in such order pre-Pandemic. And the exchange rate (KRW/EUR), consumer price index and exchange rate (KRW/USD) were ranked highly in such order during the Pandemic. Also, the lagging index, leading index and coincident index were important in such order post-Pandemic.

According to the findings of the Random Forest analysis for FSC, the coincident index was identified as significantly important post-Pandemic, unlike its lower importance pre- and during the Pandemic. Additionally, the importance of the exchange rate (KRW/EUR) increased sharply during the Pandemic.

Table 5. Findings of comparative analysis of grouping factors (Random Forest, FSC).

Classification	Pre-Pandemic		During the Pandemic		Post-Pandemic
	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank
KOSDAQ_Index	3.46%	8	3.11%	7	0.39%	10
KOSPI_Index	1.36%	9	1.21%	9	0.53%	9
Coincident_Composite_Index	1.12%	10	0.23%	10	20.48%	3
Leading_Composite_Index	14.80%	3	3.59%	5	26.40%	2
Price_Index	4.28%	7	14.40%	2	12.46%	4
W/JPY100	17.55%	2	3.59%	5	0.55%	8
W/USD	12.97%	4	4.41%	3	3.72%	5
W/CNY	11.92%	5	2.11%	8	0.57%	6
W/EUR	11.53%	6	63.50%	1	0.56%	7
Lagging_Composite_Index	21.00%	1	3.85%	4	34.34%	1

presents the findings of the comparative analysis of grouping factors (Random Forest, FSC).

4.3.3. LCC

According to the findings of the Random Forest analysis of the ratio of the number of seats supplied for international routes for LCC, the KOSPI index, KOSDAQ index and exchange rate (KRW/RMB) were ranked highly in such order pre-Pandemic. And the exchange rate (KRW/EUR), exchange rate (KRW/100 JPY) and KOSDAQ index were ranked highly in such order during the Pandemic. Also, the lagging index, exchange rate (KRW/USD) and coincident index were important in such order post-Pandemic.

According to the findings of the Random Forest analysis for LCC, the lagging index, which ranked lowest pre- and during the Pandemic, rose significantly in importance post-Pandemic. Conversely, the KOSPI index and KOSDAQ index became less important post-Pandemic, unlike their higher importance pre-Pandemic.

Table 6. Findings of comparative analysis of grouping factors (Random Forest, LCC).

Classification	Pre-Pandemic		During the Pandemic		Post-Pandemic
	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank	Feature Importance Vals (Percentage)	Rank
KOSDAQ_Index	16.93%	2	4.75%	3	0.14%	10
KOSPI_Index	17.85%	1	0.69%	5	0.16%	9
Coincident_Composite_Index	3.99%	10	0.17%	10	2.86%	3
Leading_Composite_Index	5.72%	9	0.46%	7	1.24%	5
Price_Index	12.87%	4	0.33%	8	1.86%	4
W/JPY100	6.06%	7	7.02%	2	0.25%	8
W/USD	6.14%	6	2.33%	4	8.90%	2
W/CNY	15.84%	3	0.68%	6	0.41%	6
W/EUR	6.06%	7	83.26%	1	0.29%	7
Lagging_Composite_Index	8.53%	5	0.31%	9	83.88%	1

presents the findings of the comparative analysis of grouping factors (Random Forest, LCC).

5. Discussions and Conclusions

The present research assessed various socio-economic variables related to the number of seats supplied using the SHAP and Random Forest methodologies to avoid confirmation basis and hindsight basis; identify factors influencing aircraft utilization strategies in a situation of uncertainty; uncover hidden relationships among factors; and analyze the hidden relationships based on changes in aircraft utilization rate by air route during the Pandemic. According to the analysis’ findings, pre-Pandemic, the coincident index and consumer price index were principal factors. However, the exchange rate (KRW/EUR) became important during the Pandemic. These findings underscore how the Pandemic has shifted airlines’ operational sensitivity toward macroeconomic indicators, particularly exchange rates and lagging indexes, suggesting a need for adaptive decision-making frameworks in volatile environments. For example, the European Central Bank (ECB) maintained low interest rates during the early stages of the Pandemic to ensure liquidity. However, the demand for euros has increased due to quantitative easing-related policies and a rise in interest rates in United States (ECB [52]).

According to the SHAP and Random Forest analyses, the importance of each factor is dependent on the characteristics of each methodology. The fact that an interest rate (KRW/EUR) was important in both methodologies proves that the fluctuation of interest rates had a substantial effect on the operation of airlines during the Pandemic. On the other hand, the lagging index, which ranked low pre- and during the Pandemic, rose sharply post-Pandemic. This suggests that the lagging index should be interpreted as an indicator of economic recovery specific to certain periods.

We conducted the ANOVA and Kruskal–Wallis assay to test whether the three groups differ in terms of averages and distributions. Both methodologies revealed statistically significant differences in certain cases; however, these differences were not consistently observed across all periods, particularly during and post-Pandemic. This suggests that the importance of influencing factors may vary by group and that the characteristics of each group and their respective economic environments could shape these variations.

The present research analyzed the effect of the Pandemic on aircraft utilization rate by air route and examined the importance of socio-economic variables in the strategy and operation of airlines. The SHAP and Random Forest analyses hold significance as they identify important factors to be considered for airlines to adapt to changes before, during and after periods of high uncertainty, such as the Pandemic. In particular, fluctuations in exchange rates and the period-specific business indicators emerged as critical variables for airline supply and can be regarded as essential elements in developing future airline strategies.

As a result, this research is expected to contribute to studies aimed at improving demand prediction models based on the number of seats supplied post-Pandemic, through the findings of exploratory studies including grouping factor analysis. Future studies could establish recovery and growth strategies by conducting comprehensive analyses of the various socio-economic variables and long-term time series data.

The present research analyzed factors influencing aircraft utilization using SHAP and Random Forest methodologies to minimize confirmation and hindsight biases, identify key factors shaping aircraft utilization strategies under heightened uncertainty, and reveal associations and variable importance patterns among these factors. The study also classified time into pre- and post-Pandemic periods and assessed the influence of socio-economic variables within Total, FSC, and LCC categories.

According to the analysis, it seemed that the leading index was the most important pre-Pandemic. However, the lagging index and interest rates (KRW/100 JYP, KRW/USD) became substantially more important post-Pandemic. These findings suggest that economic uncertainty and fluctuations in exchange rates had a pronounced impact on the aviation industry. In particular, the ratios of the number of seats supplied for international routes and domestic routes were reversed in FSC post-Pandemic, while the ratio of the number of seats supplied for international routes in LCC dropped to nearly 0%. This highlights differences in the recovery patterns of both models, with LLC being more severely affected during the Pandemic. According to the SHAP analysis, the coincident index and KOSDAQ index became less important pre- and post-Pandemic, whereas the lagging index rose from 10th to 1st place in terms of importance. This indicates that the lagging index, related to economic recovery, became important post-Pandemic and the fact that the lagging index, which reflects economic recovery, gained significance post-Pandemic. The growing importance of the lagging index, based on economic indicators that shift following the actual business cycle, suggests that in times of heightened market uncertainty and variability, stakeholders prioritize actual recovery indicators over predictive signals. The Random Forest analysis exhibited relatively small differences in the rankings of importance between the two methodologies.

Using Korea as case study, it was discovered that short-term international air passenger demand was influenced by economic trends and the proportion of international air passenger demand pre-Pandemic. Post-Pandemic, however, the short-term demand was influenced by artificial factors and was more sensitive to market conditions resulting from these factors than the number of confirmed cases. The short-term factors influencing the number of seats supplied for international routes, identified from a macroscopic perspective in this research, provide valuable insight for establishing future recovery strategies in the aviation industry. By examining changes in the importance of variables according to the ratio of seats supplied for international routes, this research offers a framework for anticipating shifts in airline strategies post-Pandemic. As a result, this research is expected to contribute to studies aimed at improving demand prediction models based on the number of seats supplied post-Pandemic. It is also anticipated that this research could support exploratory studies on countries beyond Korea and further studies based on their findings, ultimately aiding the development of more effective recovery strategies for the aviation industry.

The short-term factors influencing the number of seats supplied for international routes, identified from a macroscopic perspective in this research, provide valuable insight for establishing future recovery strategies in the aviation industry. By examining changes in the importance of variables according to the ratio of seats supplied for international routes, this research offers a framework for anticipating shifts in airline strategies post-Pandemic. As a result, this research is expected to contribute to studies aimed at improving demand prediction models based on the number of seats supplied post-Pandemic. It is also anticipated that this research could support exploratory studies on countries beyond Korea and further studies based on their findings, ultimately aiding the development of more effective recovery strategies for the aviation industry.