Predicting Airline Additional Services Consumption Willingness Based
on High-Dimensional Incomplete Data
S. Raja Raja Sozhan1 E. Sri Charan Reddy2
Assistant Professor B. Tech(Scholar)
Department of CSE(DS) Department of CSE(DS)
TKR College of Engineering and Technology TKR College of Engineering and Technology
Mail Id: cholan1679@gmail.com Mail Id :sricharanreddyreddy@gmail.com
B. Shashikanth Reddy3 G. Karthik4
B. Tech(Scholar) B. Tech(Scholar)
Department of CSE(DS) Department of CSE(DS)
TKR College of Engineering and Technology TKR College of Engineering and Technology
Mail Id: shashiskr36@gmail.com Mail Id :gkarthik@gmail.com
D. Sandeep5
B. Tech(Scholar)
Department of CSE(DS)
TKR College of Engineering and Technology
Mail Id: devinenisandeep0@g mail.com
ABSTRACT
Prediction of the purchase willingness of passengers has great benefits for airlines to
promote auxiliary services, however, the datasets stored in passenger travel information systems
are often high-dimensional and incomplete. This study develops a prediction method of airline
additional service consumption willingness based on high-dimensional and incomplete datasets
with a triple-layer hybrid PSO-XGBoost model, which consists of an incomplete data processing
layer, a high-dimensional data processing layer, and a predicting layer. The raw dataset is
converted into a complete and low-dimensional dataset through the first two layers and inputted
into the predicting layer to train and optimize the XGBoost model together with the PSO
algorithm and 10-fold cross-validation. The experimental results show that the proposed method
outperforms other traditional machine learning models, presenting the highest prediction score
with 0.9879 in terms of AUC. The findings help predict airline additional services consumption
intentions of passengers and are beneficial to efficient and low-cost precise marketing for airlines.
Keywords: Airfare prediction, machine learning, dynamic pricing, Random Forest, LSTM,
regression models.
The current investigation will assess the
1.Introduction different machine learning models that are being
The airline sector is now an aggressively applied in the determination of airline ticket
competitive business environment in which the prices while emphasizing at the same time how
formation of appropriate pricing methods comes these technologies play a significant role in the
across as central to revenue optimization and evolution of the aviation industry. In an attempt
survival in business. For many decades, estimate- to link theoretical innovations with practical
making for fares relied profoundly on static application, this research study will demonstrate
approaches whereby simple statistical models the applicability of machine learning to be used
were applied in using airfare histories. However, as a core component for strategic decision-
traditional approaches often fail to show the making within the airline industry.
complex dynamics of fluctuating airfare pricing.
Thus, there exists a great opportunity in machine
learning to rethink fare prediction using ever 2.Related Work
more advanced, data-centric approaches.
Airline price prediction has been regarded
as one of the foremost research areas, as it
It, therefore, allows for massive and
directly impacts the revenue management
complicated datasets to be analyzed, including
maximization and the wise passenger decision-
history fare trends, itineraries of flights,
making process. Much earlier work relied on
reservation patterns, seasonal fluctuations, and
rule-based frameworks and heuristic
broader market dynamics. Airlines utilize such
methodologies for predicting prices. The
capabilities to build models that are partially
approaches developed typically centered on
reliant on historical data while also taking into
analyzing static phenomena, such as seasonal
account real-time factors, external economic
fluctuations or peaks in demand, but were less
indicators, and competitive strategies. Such
well-suited to model dynamic factors, like
models can adapt to fluctuating pricing contexts
instantaneous booking demand or exogenous
and reveal insights that were impossible to attain
shocks (for example, weather or political events).
by traditional means.
Traditional statistical methods, like linear
regression, were also considered; however these
Grounded on such research, the
often failed to capture the nonlinear relationships
capabilities of machine learning algorithms,
typical for airfare information. The latest
which include linear regression, decision trees,
developments in machine learning are based on
random forests, and more advanced ensemble
innovative techniques around supervised models:
methods, can then be applied to improve
Decision Trees, Random Forests, and Gradient
precision and reliability in fare forecasting. The
Boosted Machines-all of which leverage
application of such advanced methodologies can
enhanced management capabilities of complex
mean optimized revenue management strategies,
datasets.
better operational effectiveness, and therefore
upgrades in customer satisfaction through
Examples include using techniques of
offering more consistent, competitive pricing.
ensemble learning for controlling variables like
day-of-week effects, booking lead time, and fare
trends over the years. These models did much
better than usual in terms of dimensions of the two elements: first, non-linear relationships
accuracy and scalability. However, the critical naturally spread in the database, and second,
weakness with these models was that they could dependencies over time. Increased accuracy of
not capture time-domain dependencies, thereby fare predictions is one of the main goals, along
leading to poor skill in predicting price with real-time analytics and sensitivity to
movements over time. More recently, deep seasonality, demand fluctuations, and
learning architectures such as Recurrent Neural competitive pricing-based strategies. A new
Networks (RNN) and Long Short-Term Memory proposed system architecture consists of three
networks were employed in addressing time- phases: data preprocessing, model training, and
domain dynamics in airline pricing. These prediction. Data pertaining to booking lead times,
architectures are known for their great ability to days of the week, and event calendars are
recognize sequential patterns in the history of cleaned, normalized, and feature-engineered for
fare data, thus enhancing predictive capacity. pre-processing from raw data. The system uses
Hybrid models which combined machine Random Forest to capture complex interactions
learning with optimization techniques, such as between the features; Graduate Boost Machines,
genetic algorithms, have also been developed to XGBoost for optimization of performance for
enhance the robustness of predictions. Despite all predictions; and LSTM networks for analyzing a
these advancements, there are considerable sequential pattern and trends over time while
challenges including real-time data integration, training models.
use of foreign variables like competitor pricing,
sparse or incomplete datasets, etc. This work
proposes a solution to all the mentioned gaps by
developing a sophisticated, machine learning-
based resilient, real-time fare prediction system.
3.Proposed System
The models run in combination to generate a
The proposed system employs machine
holistic analysis of the static and temporal price
learning methodologies to dynamically forecast
aspects. Once these models are trained and
airline ticket prices, which wipes out the
tested, the forecasting system employs real-time
limitations inherently associated with
inputs of data to project trends in airfare for the
conventional and static approaches. It is
future. The system also has a feedback
engineered to combine various sources of data,
mechanism meant to improve predictive power
including historical pricing information, real-time
further through introduction of information that
reservation data, meteorological conditions, and
is constructed from actual-new data discrepancy
important events, which makes the system
between actual and projected prices. The ultimate
provide precise and actionable forecasts.
predictions are visualized through an easy-to-use
The system in question applies the interface that enables the users to analyze fare
combination of supervised learning and deep trends and identify the best booking time.
learning strategies that can deal effectively with Moreover, the system's modular design allows
for straightforward integration with external APIs under fluctuating market conditions. All these
and data sources, ensuring scalability and limitations make it important to introduce more
flexibility in response to changing market advanced, scalable, and real-time fare-prediction
circumstances. Through integrating state-of-the- systems.
art machine learning algorithms with robust data
pipelines, the proposed system provides 4.2.Data Sources :
significant improvements both in terms of the
precision of predictions and usability regarding With regard to any airline fare prediction
airline fares predictability. model, its effectiveness greatly depends on the
quality and type of data used both in training and
forecasting. The primary sources for data about
the proposed system are:
4. Literature Survey
Historical Fare Data: This comprises historical
Airline fares have evolved from rule- pricing patterns from online travel agencies like
based systems and statistical models to advanced OTAs, airline sites, and booking platforms.
machine learning techniques. Traditional Analyzing historical data yields valuable
approaches mainly relied on historical trends, information regarding pricing patterns, seasonal
which could not suffice for the complexities of variances, and demand fluctuation, on which the
dynamic pricing. By handling nonlinear basis of developing machine learning models is
relationships, Random Forest and XGBoost derived.
delivered precision performance, and the deep
learning technique, LSTM, is a huge success in Real-time reservation information: Dynamic
capturing dependencies between time steps. variables of available tickets at present, booking
However, challenges persist with respect to rates, and competitor pricing help identify current
handling the problem of real-time data market trends and tweak predictions on-the-fly.
integration or considering effects of external APIs from the Online Travel Agencies (OTAs) or
influences; hence, adaptation and robust Global Distribution Systems (GDS) prove to be
alternatives are required. essential sources of this information.
4.1.Exsisting Systems and Their Limitations :
Extrinsic Factors: The most influential variables
on air prices will be weather, special events
Traditional airline fare forecasting approaches
(festivals, sports, concerts), and public holidays.
primarily rely on rule-based or statistical models,
Such data may be available from event calendars,
where specific algorithms are used to analyze
meteorological services, and public datasets.
past data and identify pricing patterns. Such
approaches are straightforward and relatively
easy to carry out, often focusing on variables
5.Implementation
such as lead time, seasonality, and day of the
week.
With the implementation of the proposed
Moreover, they might fail to easily combine airline fare prediction system, starting from data
with the streams of real-time data that are collection and preprocessing, stages can be
essential for precise and up-to-date predictions drawn: data is accumulated in the system from
multiple sources. These sources would involve mobile application with wide passenger
historical fare data, real-time booking APIs, and availability. Finally, the system must include a
other external factors like weather events and feedback loop for continually improving its
public holidays. The preprocessing stage predictions. During the access of real-time data,
involves the removal of duplicates and the difference between forecast and actual fare is
inconsistencies in data, normalizing data so that used to update the models at periodic intervals
it is uniform, and extraction of features such as with high precision time after time. The holistic
booking lead time, day of the week, and flight system is scaled up and designed on cloud
routes. This step will ensure consistency and platforms for all scalability factors. The system
proper structure of data under machine learning would be capable of handling vast amounts of
models that improve the system's overall data and seamlessly integrate well with external
accuracy. After data preparation, various APIs. This cloud-based approach ensures that the
machine learning models capable of forecasting system can grow and adjust itself to changing
airfare trends will be trained. market conditions, thereby propounding a
stronger solution for airline fare prediction.
It implements Random Forest Regression
for explaining complex relationships between
different features and uses XGBoost to ensure
powerful ensemble learning, thus greatly
improving predictive accuracy. Then, LSTM
Networks are employed to handle temporal
dependencies in price data efficiently to identify
sequential patterns over time. The models are
optimized in terms of hyperparameters with the
aim of maximizing their performance and indeed,
criteria through which such performance is
measured do include Mean Absolute Error
(MAE) and Root Mean Square Error (RMSE).
Core of the system is the prediction engine
combining outputs from multiple models. It uses
a weighted averaging technique such that the
resultant prediction of the three models-
combined Random Forest, XGBoost, and LSTM-
should improve in quality.
The real-time inputs such as booking data
and competitor pricing streamed through the
engine to form dynamic fare predictions aligned
with market conditions at any given time. A
subsequent user-friendly interface is produced
and includes interactive graphical representations
for users, fare trends and, finally,
recommendations on when to book a seat. This
dashboard may be developed as a website or
patterns over time. These models are also trained
on cleaned data, and strategies in hyperparameter
optimization, such as Grid Search, are added to
6.Methodology upgrade their performance. The system then puts
The proposed fare forecasting system is those models together inside a prediction engine
basically data centric and utilizes the machine that takes the output of every model. The
learning models for predicting airfare using prediction engine then combines the forecasts
historical data, current variables, and other using weighted averages so that system draws on
external influences. The first step of this process all three models to get the most accurate possible
involves data collection, which could involve prediction. The system is also designed to
gathering several kinds of data streams such as adaptively introduce actual current inputs, so that
historical pricing information, booking details, the fare forecast could be modified based on
meteorological conditions, and major events. market conditions prevailing at the time of travel,
These sources of data are important for such as demand and competitive pricing
understanding the variables affecting the fair strategies. Finally, a feedback loop is introduced
variation and provide a comprehensive dataset to automatically and continuously refine the
for the training of models. Later, the data enters accuracy of the system. Feedback is collected in
processes such as cleaning, normalization, and the form of real-time discrepancies between
preprocessing to ensure consistency and remove forecast and actual fares for models to
potential noise or outliers that may impact the periodically retrain, ensuring continued
performance of the model. On this dataset, effectiveness over the long term. This approach
feature engineering is performed to generate guarantees that the airline fare prediction system
critical variables that involve booking lead time, remains both responsive and precise, effectively
seasonal trends, flight routes, and demand responding to the fluid characteristics of airline
patterns-all of which are pivotal in an exact pricing.
airline fare price forecast. The airfare is then
predicted after the cleaning process by applying 7.Future Scope
machine learning models.
The projected potential of the fare
The system utilizes a variety of prediction system proposed seems to be quite
algorithms to maintain both robustness and optimistic and promising, offering many scopes
accuracy. Random Forest Regression is for development and extension. One of these
implemented to identify non-linear relationships conceivable possibilities is the use of real-time
among multiple features, thereby facilitating sensor data and wearable technology to track
dependable fare predictions. passenger behavior and preferences, such as
journey patterns, traveling expenditure, or even
XGBoost is employed as it allows for the
emotional state while making a reservation. This
improvement of model performances by
dataset could lead to an enhancement in the
boosting-based approaches, which helps in
personalization of fare predictions, allowing
increasing the accuracy in predictions. In
airlines to create dynamic pricing approaches
addition to this, Long Short-Term Memory
tailored not only to prevailing market conditions
(LSTM) networks are also added to tackle the
but also to individual customer behaviors and
temporal dependency shown within the data-sets
preferences. The addition of AI-based predictive
for the model to identify the fare variation
analytics may further enable the system to predict advanced machine learning methodologies.
shifts in demand resulting from unanticipated Using an integration of historical data, real-time
factors such as economic changeovers, natural variables, and external influences, this system is
disasters, or surprise market shocks that would able to produce forecasts on the actual
allow for more accurate and proactive implementable level of airline prices, enabling
adjustments in fares. both air travelers and airlines alike to fine-tune
their decisions. The use of models such as
Another area of improvement is in the Random Forest, XGBoost, and LSTM enables
integration of multi-modal transport data. In the system to identify non-linear associations as
doing so, an integrated travel solution could well as temporal dependencies, resulting in
potentially be offered by the fare prediction reliable predictions in a complex and dynamic
model, thereby making it highly useful to market environment.
individuals requiring multi-mode transportation
on a trip. It will primarily be helpful for those Although the system has achieved
planning complicated itineraries involving progress, it becomes a flexible tool, able to keep
various modes of transport. In this regard, the abreast with change in respect to real-time
adaptability reinforced by a feedback mechanism
system could make more accurate predictions
for continuity in learning. Those external
concerning overall travel costs while aiding the variables - meteorological activity, demand
user in making better decisions pertaining to variability, competitors' pricing models - help it
planning their entire journey. refine the predictive model further. Predicting
and analyzing future changes, several
Finally, international expansion may possibilities now open for improving the
make the system more inclusive and relevant functions of this system by including individual
across various markets and regions. Integration fare forecasts, diverse types of transport, and
of multilingual support and adaptation to other international adaptability by making it a whole-
cultural frameworks can make the system more package solution for today's traveler and for the
inclusive and relevant to a larger group, so that aviation industry as well. With this framework,
customized fair predictions can be achieved for there is the prospect of transforming fare
international travelers. Further development in forecasting to be more dependable and
machine learning and big data analytics may customized to add value to either consumers or
eventually enable the system to handle larger stakeholders in the aviation sector.
datasets, increase precision, and scale
appropriately. These developments might render
the fare prediction system an all-round resource 9.References
not only for travelers, but also for the airlines,
travel agencies, and policymakers who are [1] Jordan, M. I., & Mitchell, T. M. (2015).
concerned with enhancing pricing strategies and Machine learning: Trends, perspectives, and
improving customer satisfaction among people prospects. Science, 349(6245), 255-260.
across the globe.
[2] Hastie, T., Tibshirani, R., Friedman, J. H., &
8.Conclusion Friedman, J. H. (2009). The elements of
The proposed system for predicting statistical learning: data mining, inference,
airline fare suggests a very significant and prediction (Vol. 2, pp. 1-758). New
advancement in the field of dynamic pricing and York: springer.
travel arrangements, as it is powered by
[3] Valiant, L. G. (1984). A theory of the Predicting Local Airfare Prices with Deep
learnable. Communications of the ACM, Transfer Learning Technique. In 2023
27(11), 1134-1142. Innovations in Intelligent Systems and
[4] Rao, N. S. S. V. S., & Thangaraj, S. J. J. Applications Conference (ASYU) (pp. 1-4).
(2023, [10] Malkawi,M.,&Alhajj, R.(2023,
April). Flight Ticket Prediction using Random August).Real-timeweb-basedInternational
Forest Regressor Compared with Decision Flight TicketsRecommendation System
Tree Regressor. In 2023 Eighth International via Apache Spark. In 2023 IEEE 24th
Conference on Science Technology International Conference on Information
Engineering and Mathematics (ICONSTEM) Reuse and Integration for Data Science
(pp. 1-5). IEEE. (IRI) (pp. 279-282). IEEE.
[11] Joshitta, S. M., Sunil, M. P., Bodhankar, A.,
[5] Burger, B., & Fuchs, M. (2005). Dynamic
Sreedevi, C., & Khanna, R. (2023, May).
pricing—A future airline business model. The Integration of Machine Learning
Journal of Revenue and Pricing Management, Technique with the Existing System to
4(1), 39-53. Predict the Flight Prices. In 2023 3rd
[6] Malighetti, P., Paleari, S., & Redondi, R. [12] International Conference on Advance
(2010). Has Ryanair's pricing strategy Computing and Innovative Technologies in
changed over time? An empirical analysis of Engineering (ICACITE) (pp. 398-402).
its 2006–2007 flights. Tourism management, IEEE.
31(1), 36-44.
[13] Groves, W., & Gini, M. (2013, May). An
[7] Liu, T., Cao, J., Tan, Y., & Xiao, Q. (2017, agent for optimizing airline ticket
December). ACER: An adaptive context- purchasing. In Proceedings of the 2013
international conference on Autonomous
aware ensemble regression model for airfare
agents and multi-agent systems (pp. 1341-
price prediction. In 2017 International
1342).
Conference on Progress in Informatics and
Computing (PIC) (pp. 312-317). IEEE.
[14] Domínguez-Menchero, J. S., Rivera, J., &
Torres-Manzanera, E. (2014). Optimal
[8] Tziridis, K., Kalampokas, T., Papakostas, G.
purchase timing in the airline market.
A., & Diamantaras, K. I. (2017, August). Journal of Air Transport Management, 40,
Airfare prices prediction using machine 137-143.
learning techniques. In 2017 25th European
Signal Processing Conference (EUSIPCO) [15] Sarao, P., & Samanta, P. (2022). Flight Fare
(pp. 1036-1039). IEEE. Prediction Using Machine Learning.
Available at SSRN 4269263.
[9] Can, Y. S., & Alagöz, F. (2023, October).
[16] Python Software Foundation. Python 3
Documentation: https://docs.python.org/3/ - manipulation library, was used extensively for
The official documentation for Python 3, data preprocessing, cleaning, and analysis in
pivotal in the development and
the flight price prediction project.
implementation of machine learning models
for flight price prediction.
[19] Scikit-learnDevelopers.Scikit-learn
[17] NumPyCommunity. NumPy Documentation: https://scikitlearn.org/stable/
Documentation: documentation.html Scikit-learn provided a
wide range of machine learning tools and
https://numpy.org/doc/stable/ - NumPy, a
models, including Random Forest, employed
fundamental package for scientific
for building predictive models in the project.
computing with Python, played a crucial role
in handling numerical operations and data [20] Matplotlib Development Team. Matplotlib
manipulation in the flight price prediction Documentation:https://matplotlib.org/
project. stable/contents.html Matplotlib, a popular
data visualization library, played a key role
in creating visualizations and plots to
[18]Pandas Development Team. Pandas
analyze
Documentation:https://pandas.pydata.org/pand
trends and patterns in flight price data.
asdocs/stable/ Pandas, a powerful data
