Predictive Analytics for Enhanced Passenger Satisfaction in the Airline Industry:

Leveraging Machine Learning to Drive Strategic Decision-Making

Walaa Saber Ismail Hesham Bawazeer Hamad Almansori

Business Information Technology dept. Business Information Technology dept. Business Information Technology dept.
Liwa College Liwa College Liwa College
Abu Dhabi, UAE Abu Dhabi, UAE Abu Dhabi, UAE
walaa.saber@lc.ac.ae hesham012414@lc.ac.ae hamad013486@lc.ac.ae
2024 10th International Conference on Optimization and Applications (ICOA) | 979-8-3503-8735-3/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICOA62581.2024.10753807

Abstract __ In the highly competitive airline industry, enhancing

passenger satisfaction is crucial for maintaining customer loyalty and
By incorporating advanced machine learning techniques, this study
market competitiveness. This study utilizes advanced machine learning not only identifies key factors but also provides actionable
techniques to analyze comprehensive dataset comprising recommendations for service enhancement.
approximately 130,000 entries related to airline passenger experiences.
The dataset includes both qualitative and quantitative attributes from II. RELATED WORK
in-flight services to logistical details of air travel. Two predominant
machine learning models, Naive Bayes and k-Nearest Neighbors (k- In recent years, the aviation industry has increasingly focused on
NN) were employed to predict passenger satisfaction levels and identify leveraging predictive analytics to enhance passenger satisfaction.
key factors impacting their experience. The Naive Bayes model This literature review summarizes key studies and methodologies
demonstrated superior predictive performance compared to k-NN, that underpin this research domain, emphasizing machine learning
with significant accuracy, precision, recall, and AUC metrics. This
research not only offers valuable insights into the key factors
applications for improving airline service quality and customer
influencing passenger satisfaction but also provides data-driven experiences [6].
recommendations for service improvement. It lays the groundwork for
airlines to utilize predictive analytics in strategic decision-making, Several studies have explored the application of machine learning
thereby enhancing passenger satisfaction and operational efficiency. models to predict and improve passenger satisfaction in the airline
industry. The integration of predictive analytics into service quality
Keywords – Airline industry, passenger satisfaction, machine
assessments is seen as a strategic tool for gaining competitive
learning, predictive analytics, Naive Bayes, k-NN, in-flight services,
advantage and customer loyalty. For instance, Park et al.,
service improvement.
demonstrated that data-driven insights could significantly improve
I. INTRODUCTION operational efficiencies and passenger satisfaction levels by
tailoring services to meet customer expectations [7].
In the aviation industry, which is continuous and progressive in
nature, maintaining and enhancing passenger satisfaction is crucial Among various machine learning models, Naive Bayes and k-
for customer loyalty and competitive advantage. In this context, Nearest Neighbors (k-NN) have been extensively studied for their
understanding the multifaceted elements that influence a passenger's effectiveness in classification tasks related to customer satisfaction
overall satisfaction with their flight experience becomes imperative. [3]. Naive Bayes, known for its simplicity and effectiveness in high-
This study aims to investigate various factors that contribute to dimensional datasets, has been shown to perform well in predicting
passenger satisfaction, leveraging data analytics and machine categorical outcomes such as satisfaction levels [2]. Jiang, Zhang,
learning to derive meaningful insights [1]. and Su conducted an empirical study that highlighted the robustness
of Naive Bayes in handling different data types and its applicability
Machine learning plays a pivotal role in analyzing large volumes of in real-world scenarios [8].
data to uncover patterns and trends that might not be immediately
apparent through traditional analytical methods. By employing On the other hand, k-NN is valued for its straightforward
models such as Naive Bayes [2] and K-Nearest Neighbor [3] implementation and interpretability. Kang and Cho discussed the
classifiers, the study classifies passenger satisfaction levels based advantages of k-NN in terms of its instance-based learning
on numerous service attributes, highlighting areas that are most approach, which can be particularly effective in scenarios where the
influential in shaping their overall experience. This approach relationship between variables is non-linear and complex. However,
ensures a robust and comprehensive understanding of the factors k-NN's performance is highly dependent on the choice of k and the
driving passenger satisfaction, enabling airlines to implement data- distance metric used, necessitating careful tuning for optimal results
driven improvements in their services [4]. [9].

The development of a comprehensive data framework in this study Research has identified several key factors that significantly impact
includes diverse variables such as in-flight services and logistical airline passenger satisfaction. These include service quality
aspects of travel, measuring their impact on passenger satisfaction. attributes such as seat comfort, in-flight entertainment, and the
Previous research has shown that service quality directly influences efficiency of check-in services. A study by Walia, Sharma, and
passenger satisfaction and loyalty [5]. Mathur revealed that tangible and intangible service aspects
979-8-3503-8735-3/24/$31.00 ©2024
Authorized licensed use limited to: De Montfort IEEE
University. Downloaded on February 10,2025 at 03:47:19 UTC from IEEE Xplore. Restrictions apply.
collectively shape the overall passenger experience. Their findings RapidMiner software, which provides robust tools for data
suggest that airlines must focus on both operational efficiencies and preprocessing, visualization, and model development [13].
customer service improvements to enhance satisfaction levels [10].
III. RESEARCH METHODOLOGY
Moreover, studies like that of Archana and Subha emphasize the
importance of addressing delays and improving punctuality as A. Dataset Understanding
critical factors influencing passenger satisfaction. Their research This study uses a dataset called "Airline Passenger Satisfaction" to
indicates that reducing both departure and arrival delays can analyze factors affecting passenger satisfaction across various
substantially enhance customer perceptions of airline reliability and airline services. The dataset, which includes 130,000 entries,
service quality [11]. includes 23 features such as demographic information, service
ratings, and flight details. Key features include qualitative
The application of predictive analytics extends beyond mere assessments like inflight wifi, food and drink quality, seat comfort,
prediction to provide actionable recommendations for service and overall satisfaction. Quantitative data includes flight distance
improvement. By analyzing comprehensive datasets that include and delay times. The dataset, originally sourced from Kaggle, has
both qualitative and quantitative attributes of passenger been cleaned and preprocessed for accuracy and relevance for
experiences, airlines can identify specific areas for enhancement. predictive modeling. It provides a multifaceted view of the customer
For example, Kobi and Otieno demonstrated that machine learning base and service-specific ratings. The comprehensive nature of the
models could uncover hidden patterns in passenger feedback, data makes it a valuable resource for enhancing passenger
enabling airlines to make informed decisions about service satisfaction and service efficiency [14].
modifications [12].
B. Data Pre-processing
Despite these advancements, challenges remain in ensuring data To prepare the dataset for further analysis we used following steps
quality, handling imbalanced datasets, and addressing privacy (see Fig. 1) using Rapidminer [15], which are described below:
concerns related to the use of passenger data. Continuous innovation i. Importing Dataset
and adherence to ethical standards will be crucial in leveraging
There two "Read CSV" operators are used to import raw data,
predictive analytics to its full potential in the airline industry.
(training and testing datasets are in separate files) for accurate
model validation and comprehensive features of flight
A. Objectives: experience and customer demographics.
The primary objectives of this study are as follows: ii. Attributes Selection
• Predictive Analysis: Develop predictive models to accurately
The "Select Attributes" operators were used to select attributes
determine passenger satisfaction levels based on various
for further analysis, excluding 'att1' and 'id' due to their
factors.
irrelevantness and lack of predictive power.
• Service Improvement Insights: Identify key areas for
improvement in airline services to enhance passenger iii. Handling Missing Values
satisfaction. The "Replace Missing Values" operators address missing
• Strategic Recommendations: Provide actionable values in predictive models by replacing them with the average
recommendations for optimizing airline services to improve of their respective columns, assuming randomness and
customer satisfaction and increase profitability. minimal skew.
iv. Assigning Role
The study will utilize a comprehensive dataset sourced from Kaggle,
The "Set Role" operator assigns a specific role to an attribute
which includes detailed passenger feedback on various aspects of
in supervised learning tasks, defining the target variable as
their flight experience. The dataset will be analyzed using
"Satisfaction" and the rest as independent variables.

Figure 1: Data Preparation Rapidminer Process

C. Predictive Analysis of 5 chosen to balance noise and accuracy. The choice of K can be
The study uses two machine learning algorithms, K-Nearest refined based on cross-validation results. Naive Bayes is a
Neighbors (KNN) [3] and Naive Bayes [2], to predict airline probabilistic classifier that uses Laplace correction to handle zero
passenger satisfaction levels. KNN is a lazy learning algorithm that probabilities in the dataset. By adjusting probability estimates to be
classifies new cases based on similarity measures, with a K value more robust to unseen data, the model becomes more
generalizable.

Authorized licensed use limited to: De Montfort University. Downloaded on February 10,2025 at 03:47:19 UTC from IEEE Xplore. Restrictions apply.
 D. Model’s Evaluation predictive abilities. Precision measures the correctness of the
The "Performance" operator evaluates the models' predictions by positive predictions, while recall, also known as sensitivity,
comparing them to the actual outcomes using various metrics. measures the model's ability to identify all actual positives. Other
Accuracy is the most straightforward performance measure, but metrics, such as the F1-score, may be considered to balance
precision and recall provide deeper insights into the models' precision and recall, particularly if there is a class imbalance in the
dataset [16].

Figure 2: Rapidminer Process for complete Analysis

IV. INSIGHTS
The airline passenger satisfaction dataset was analyzed using
RapidMiner's built-in visualization capabilities to identify patterns and
correlations that could impact passenger satisfaction.
• Age Distribution and Satisfaction
Histograms segmented by satisfaction level showed (see Fig. 3)
that younger and older passengers are more dissatisfied than
middle-aged passengers.

Figure 3: : Satisfaction distribution across different ages

• Service Ratings vs. Satisfaction

Box plots(see Fig. 4) revealed a strong correlation between higher
satisfaction ratings in seat comfort and inflight entertainment and
higher satisfaction levels across various service features.

Figure 4: Inflight Service Ratings vs. Satisfaction distribution using box plot

• Flight Distance and Class Longer flights in business class showed higher satisfaction
We plotted (see Fig. 5) flight distance against travel class to rates, indicating that comfort becomes more critical on longer
examine trends in passenger distribution and satisfaction. journeys.

Authorized licensed use limited to: De Montfort University. Downloaded on February 10,2025 at 03:47:19 UTC from IEEE Xplore. Restrictions apply.
 Figure 5: Flight Distance and Class distribution using Bar chart

• Delay Impact associated with higher satisfaction. This visualization

Scatter plots (see Fig. 6) of departure and arrival delays also showed that dissatisfaction increases sharply with
against satisfaction highlighted that shorter delays are delay duration as you see in.

Figure 6: Scatter Plot visualization between Departure Delay and Satisfaction

• Correlation Matrix correlations between on-board and inflight service, both
A heatmap identifying relationships between numerical correlated with overall satisfaction.
variables (see Fig. 7) revealed significant positive

Figure 7: Heatmap visualization for dataset

V. RESULTS A. K-Nearest Neighbors (KNN)

The performance metrics of K-Nearest Neighbors and Naive The KNN model, with a k=5 setting, accurately predicted
Bayes models on the airline passenger satisfaction dataset reveal satisfaction levels for 74.52% of passengers in a dataset. It
their predictive capabilities. was particularly precise in identifying satisfied customers,
with a precision rate of 73%. However, its recall rate was

Authorized licensed use limited to: De Montfort University. Downloaded on February 10,2025 at 03:47:19 UTC from IEEE Xplore. Restrictions apply.
 lower at 66.59%, indicating it missed some satisfied
passengers. The model's AUC was solid, averaging at 0.810.

Figure 8: KNN Model Performance

(0.912), indicating its superior ability to accurately classify
B. Naïve Bayes passengers' satisfaction, identify truly satisfied passengers,
The Naive Bayes model outperformed KNN in key and differentiate between satisfied and less satisfied
performance metrics, with higher accuracy (84.46%), passengers.
precision (82.24%), recall rate (82.39%), and excellent AUC

Figure 9: Naive Bayes Model Performance

The Naive Bayes model demonstrated superior accuracy and has a higher AUC value, indicating its superior predictive power.
reliability in predicting passenger satisfaction, outperforming the This highlights its superior ability to distinguish between classes.
KNN model in terms of accuracy, precision, recall, and AUC in
this dataset.
VI. Performance Comparison
Here's the comparison table (see tab. 1) showing the performance
metrics of the K-Nearest Neighbors (KNN) and Naive Bayes
models:

Table 1: Performance Comparison Table for both Models

METRIC KNN (%) NAIVE BAYES (%)
Accuracy 74.52 84.46
Precision 73.00 82.24
Recall 66.59 82.39
AUC 0.800 - 0.820 0.912

The Naive Bayes model outperforms the KNN model in predicting

airline passenger satisfaction, with higher accuracy, precision, and Figure 10: Both Classifiers performance Comparison using Bar Chart
recall. It is more reliable in identifying satisfied passengers and
VII. CONCLUSION AND FUTURE WORK Looking forward, the study can be extended in several ways.
The analysis conducted in this study has provided valuable Integrating additional data sources like real-time feedback via
insights into the factors that significantly influence airline social media could enrich the analysis and provide a more nuanced
passenger satisfaction. Utilizing advanced machine learning understanding of satisfaction drivers. Developing systems for real-
techniques, specifically Naive Bayes and K-Nearest Neighbors, time analytics could enable airlines to address service issues more
we identified critical service quality attributes such as seat promptly, potentially transforming customer experiences
comfort, inflight entertainment, and the efficiency of check-in dynamically. Additionally, applying personalization algorithms to
services that play pivotal roles in shaping passenger experiences. tailor services based on individual preferences could further
Among the models tested, Naive Bayes exhibited superior enhance passenger satisfaction and provide a competitive edge in
performance, with higher accuracy and better overall predictive the market. Experimenting with more advanced models and
capabilities, suggesting it is particularly suited to handling the ensemble methods could also improve the robustness and
complex data types involved in passenger feedback. accuracy of predictions, while conducting longitudinal studies

Authorized licensed use limited to: De Montfort University. Downloaded on February 10,2025 at 03:47:19 UTC from IEEE Xplore. Restrictions apply.
could help track changes in passenger expectations over time, [13]. M. S. Mousavian, S. Miah, and Y. Zhong, "A design
allowing airlines to adapt their strategies to shifting market concept of big data analytics model for managers in
conditions. By pursuing these avenues, airlines can refine their hospitality industries," Personal and Ubiquitous
approaches to improving passenger satisfaction, ensuring that Computing, vol. 27, pp. 1-11, 2023, doi: 10.1007/s00779-
their service enhancements are both strategic and effective. 023-01714-3.
[14]. J. D., "Passenger Satisfaction," 2018. [Online].
VIII. REFERENCES Available:
https://www.kaggle.com/datasets/teejmahal20/airline-
[1]. M. Salamoura, I. Chaniotakis, and C. Lymperopoulos, passenger-satisfaction?select=train.csv.
"Enhancing Airline Passengers' Satisfaction Through [15]. Mierswa and R. Klinkenberg, "RapidMiner Studio
Service Quality: The Importance of the Human Factor," (9.1)," Data science, machine learning, predictive
Journal of Air Transport Studies, vol. 8, pp. 54-69, 2017, analytics, 2018. [Online]. Available:
doi: 10.38008/jats.v8i2.32. https://rapidminer.com/.
[2]. Rish, "An Empirical Study of the Naïve Bayes [16]. Y. Shi, "A Machine Learning Study on the Model
Classifier," in IJCAI 2001 Work Empir Methods Artif Performance of Human Resources Predictive Algorithms,"
Intell., vol. 3, 2001. in 2022 4th International Conference on Applied Machine
[3]. G. Guo, H. Wang, D. Bell, and Y. Bi, "KNN Model- Learning (ICAML), Changsha, China, 2022, pp. 405-409,
Based Approach in Classification," 2004. doi: 10.1109/ICAML57167.2022.00082.
[4]. S. Mohamed and H. Klaus, "AI and ML for Data-Driven
Insights: Machine learning algorithms can analyze vast
amounts of medical data to identify patterns and trends,"
2024.
[5]. G. C. Saha and Theingi, "Service quality, satisfaction,
and behavioural intentions: A study of low-cost airline
carriers in Thailand," Managing Service Quality: An
International Journal, vol. 19, no. 3, pp. 350-372, 2009.
[6]. M. Rampersad-Jagmohan and Y. Wang, "Predictive
Analytics in Aviation Management," 2024, doi:
10.1007/978-981-97-0665-5_52.
[7]. M.-Y. Park, Y.-J. Kim, and Y.-H. Park, "A Deep
Learning Approach to Analyze Airline Customer
Propensities: The Case of South Korea," Applied
Sciences, vol. 12, no. 4, p. 1916, 2022, doi:
10.3390/app12041916.
[8]. L. Jiang, H. Zhang, and J. Su, "Learning k-Nearest
Neighbor Naive Bayes for Ranking," in Lecture Notes in
Computer Science, vol. 3584, pp. 175-185, 2005, doi:
10.1007/11527503_21.
[9]. P. Kang and S. Cho, "Locally linear reconstruction for
instance-based learning," Pattern Recognition, vol. 41,
no. 11, pp. 3507-3518, 2008, doi:
10.1016/j.patcog.2008.04.009.
[10]. S. Walia, D. Sharma, and A. Mathur, "The Impact of
Service Quality on Passenger Satisfaction and Loyalty in
the Indian Aviation Industry," 2021.
[11]. R. Archana and M. Subha, "A study on service quality
and passenger satisfaction on Indian airlines,"
International Journal of Multidisciplinary Research, vol.
2, 2012.
[12]. J. Kobi and B. Otieno, "Predictive Analytics
Applications for Enhanced Customer Retention and
Increased Profitability in the Telecommunications
Industry," International Journal of Innovative Science and
Research Technology, vol. 9, 2024, doi:
10.38124/ijisrt/IJISRT24MAY1148.

Authorized licensed use limited to: De Montfort University. Downloaded on February 10,2025 at 03:47:19 UTC from IEEE Xplore. Restrictions apply.