Data-Driven Insights for Improving Airline Passenger Experience

Shaoqing Zhang Dev Kotak Yu Di

Sprott School of Business Department of Systems and Computer Sprott School of Business
Introduction Visulisation Engineering Result
As the aviation industry approaches a post-pandemic • Our investigation set out on a mission to understand the complexities of airline customer happiness using a
resurgence, with passenger traffic projected to fully thorough data-driven methodology. We aimed to identify the aspects that have the greatest impact on the
recover by 2025, the aim to understand and improve
Results
passenger experience by using the capabilities of several machine learning models.
airline passenger satisfaction becomes paramount.
This project analyzes a comprehensive dataset to explore • We observed that in-flight entertainment, seat comfort, and the simplicity of online booking were the most
the multifaceted nature of passenger experiences, important factors in passenger satisfaction. The investigation included five important models: Logistic
leveraging statistical and machine learning techniques to Regression, Decision Tree, Random Forest, XGBoost, and SVM. Among them, Logistic Regression and SVM
identify key satisfaction drivers and predict future trends. gave basic insights, with accuracies of 89.35% and 89.47%, respectively. However, it was the Random Forest
Among the challenges and transformations facing and XGBoost models that really shined, with each achieving an outstanding accuracy of 96.66%,
airlines, this data-driven analysis aims to offer actionable demonstrating their effectiveness in projecting passenger happiness. The Decision Tree model also
insights and recommendations, fostering improved performed well, with an accuracy of 94.42%, highlighting the usefulness of these analytical techniques in
service quality, customer loyalty, and a deeper traversing the complexity of passenger preferences.
understanding of the complex dynamics shaping
passenger satisfaction in the evolving landscape of air • Our results not only shed light on the primary causes of customer happiness, but also provided a roadmap
travel. Figure:1 Data Distribution across features for airlines to improve their services. Airlines may boost their service offerings by concentrating on in-flight
It equips industry stakeholders with the tools for strategic Figure 1 presents a composite view of airline passenger entertainment, seat comfort, and streamlining the online booking process, promoting an atmosphere of
decision-making, ensuring a responsive and customer- satisfaction metrics, with color coding separating satisfied increased customer loyalty and pleasure.
centric approach to service enhancement. customers from neutral or dissatisfied ones.
The visuals explore correlations between satisfaction and • The accompanying bar graph clearly depicts each model's performance, allowing for a clear visual
factors like flight distance, cleanliness, and type of travel. comparison that highlights the Random Forest and XGBoost models' greater accuracy. This graphical
Datasets and Limitations The distribution of satisfaction across classes and the depiction demonstrates the rigorous analytical approach used, emphasizing the strategic significance of our
demographics of travelers further inform on the results for the airline sector. Through this endeavour, we want to contribute to continuous efforts to improve
This analysis is anchored in a dataset derived from a
underlying patterns, offering airlines actionable insights for the air travel experience, ensuring that customer pleasure stays at the top of the list of industry objectives.
passenger satisfaction dataset, thoroughly compiled to
improving passenger experience.
explore the myriad factors contributing to passenger
satisfaction or dissatisfaction.
The dataset encapsulates a broad spectrum of variables, • The confusion matrix displayed in Figure 3 reveals a Conclusion
including demographic information (Gender, Age, high number of true positives and true negatives,
• Our project successfully extracted key satisfaction
Customer Type), travel specifics (Type of Travel, Class, suggesting accurate predictions of customer
satisfaction levels. It also shows relatively few false drivers from airline passenger surveys, highlighting
Flight Distance), and detailed satisfaction ratings on inflight entertainment, seat comfort, and online
various service dimensions (Inflight wi-fi, Online positives and false negatives, indicating rare instances
where predictions of customer satisfaction were booking as crucial to enhancing passenger
Booking, Seat Comfort, Cleanliness, etc.).
incorrect. This model's predictive accuracy is evident experience. Through advanced data analytics, we
However, this investigation is not without limitations.
in its ability to discern between satisfied and pinpointed areas for improvement, notably in off-
1. There are missing values in the "Arrival Delay in
dissatisfied clients effectively. board services. Our models, especially the decision
Minutes" column of the training dataset, which could
tree, proved effective in accurately predicting
affect analyses related to flight delays and passenger
satisfaction. passenger sentiment, providing a roadmap for
2. The dataset includes several subjective measures airlines to foster loyalty and refine service quality.
of satisfaction (e.g., seat comfort, cleanliness) rated on
a scale. These subjective ratings can vary widely
Future Work
among passengers based on individual expectations
and experiences. • In future studies, we plan to compare the
performance of machine learning models trained
on different datasets to assess the impact of data
variability on model accuracy.
• Additionally, we will investigate the importance of
Figure:2 Feature Importance graph different features in predicting diabetes risk and
compare their performance across different
The bar graph illustrates the importance of various
3.The data is based on passenger surveys, which are models. This analysis will provide valuable insights
features in determining airline passenger satisfaction.
inherently subject to biases such as response bias, into the factors that contribute to diabetes risk and
Online booking emerges as the most significant factor,
where individuals might report higher or lower enable the development of more accurate and
followed by inflight wi-fi service and type of travel.
satisfaction levels due to perceived social desirability or reliable machine learning models for diabetes risk
These insights could guide airlines in prioritizing Figure:5 Model Accuracy Comparison
personal biases.
enhancements to their booking systems, inflight prediction.
Methodology connectivity, and tailored services for different travel • The image displays a vertical bar graph comparing
types to boost customer satisfaction. References
the accuracy scores of five different machine
Data Loading & Exploration learning models. Each bar, colored uniquely, 1. An Analysis Of Airlines Customer Satisfaction By Improving
Customer Service Performance by kevin Tegar s et. al.
Initial exploration involves checking data types, unique represents a model: Logistic Regression, Decision January 2018.
values, and counts of labels. Tree, Random Forest, XGBoost, and SVM. The
accuracy scores are marked on top of each bar, with 2. Airline satisfaction and loyalty: Assessing the influence of
Data Cleaning &Preparation Random Forest and XGBoost sharing the highest personality, trust and service quality by Steven Leon et. al
Steps include handling missing values, encoding score at approximately 96.66%, followed by Decision October 2023.
categorical variables, and splitting the data into training Tree, SVM, and Logistic Regression. The graph is
and testing sets. well-organized with clear labels for both the axes 3. Customer Satisfaction with Commercial Airlines: The Role of
and the bars, facilitating easy comparison of the Perceived Safety and Purpose of Travel by Christian M.
ML Model Building models' performances.
Ringle et. al .OCtober 2011.
Various classifiers are instantiated and trained on the • The color coding enhances the visual distinction
training data. Predictions are made on the test set. between the models, making the information Acknowledgement
Evaluation accessible even at a cursory view. This graph serves We would like to extend out thank you to Prof. Majid Komeili
The model's effectiveness is quantified by key metrics as an effective summary for analyzing the predictive from The Carleton University for his support, guidance, and
and a confusion matrix for a complete performance strength of the various models presented. knowledge.
Figure:3 Confusion Matrix
assessment.