Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14944858

Dynamic Airline Pricing System

Tilak Bhujade1; Harshal Hingankar2; Janvi Charde3; Dr. Tejal Irkhede4
1;2;3;4
Under the Guidance of Computer Science and Engineering (Data Science),
St. Vincent Pallotti College of Engineering and Technology, Nagpur 441108

Publication Date: 2025/02/28

Abstract: Competition over fare control has reached a new level of complexity in the airline industry through machine
learning to determine the most effective ticket pricing strategies. This research paper demonstrates an ideal dynamic
pricing model developed on the programming language Python including preprocessing of data, selection of features and
other state of the art models Random Forest and Prophet model among others. The model takes data flights details,
economic conditions, weather conditions, and customers demographics of the flight to predict ticket prices correctly. Due
to the interface, implemented with Streamlit, the model enables users to input numerous parameters and obtain flight
price estimations. The results clearly bring out the possibility of the use of machine learning in the airline industries to
improve the revenue management and therefore increase the right price solution and customer satisfaction. This paper
seeks to add on the existing literature pertaining to ERP and Advance Metering Infrastructure and its implementation in
Airline Industry.

How to Cite: Tilak Bhujade; Harshal Hingankar; Janvi Charde; Dr. Tejal Irkhede (2025). Dynamic Airline Pricing System.
International Journal of Innovative Science and Research Technology, 10 (1), 2900-2907.
https://doi.org/10.5281/zenodo.14944858

I. INTRODUCTION with past data and trying to predict trends that are
forthcoming. Because these models can recognize subtleties
A relatively recent growth to the system of airline fare that may not be easily seen by first glance, airlines can make
is dynamic scheduling that has enabled airlines to change sound price determination. Moreover, the current study
ticket prices in real time depending on the prevailing establishes comprehensible interfaces to enhance the basic
demand in the market, as well as pricing offered by other interactions with the model by stakeholders to perform
airlines and other factors. This form of pricing is different additional analyses for other scenarios and dynamic
from the fixed pricing strategic models successfully adjustments of the prices by the concerned users provided
implemented by many airlines, thus allowing flexible and they have set up parameters in their preferred formats.
responsive pricing for the optimization of revenues. The
possibility to vary them can not only increase profits but So, as the airline industry grows, the role of dynamic
also guarantee customers’ appreciation, offering favourable pricing will only rise to meet it. The application of artificial
offerings that correspond to their needs and available offers intelligence and machine learning in the pricing not only
at a certain point in time. defines a competitive advantage but also matches the trends
in the increased focus on the individual approach to
The high levels of heterogeneity in the consumers’ traveling. The purpose of this research paper is to present
behaviour and the overall demand make it important for detailed information regarding dynamic airline pricing and
firms to develop complex models for pricing that can critical analysis of the selected machine learning-based
accommodate this information. Airlines ingest data from a model used to explain the nature and major aspects of
number of sources such as travel patterns, occasions, pricing models in the airline industry.
economic variables, and even climate. Through the
establishment of such interface, the individual values of In this paper, we present before you our “Dynamic
these databases can be consolidated into a meaningful model Airline Pricing System”, which is an innovative solution to a
that helps airlines forecast passage transactions and set classical problem of static pricing.
realistic prices based on current market conditions. It also
proves useful in working towards efficient capacity control, II. LITERATURE REVIEW
and filling the airplane to its capacity as much as possible.
Continuous pricing techniques have drawn much
Advanced approaches to intelligent computation are attention in airline sectors, where they are applied to
particularly critical in the work on dynamic pricing models. maximize revenues through variable pricing of the tickets.
Techniques such as Random Forest and Prophet are used Chen et al. (2016) [1] has defined dynamic pricing as a

IJISRT25JAN1558 www.ijisrt.com 2900

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14944858
strategy that enhances the airlines’ ability to seize the such as target price with reference to the overall customer
changing opportunities in the market as it occurs. The segmentation the level of customer satisfaction and loyalty
authors point out that with incorporation of advanced can be enhanced together with the corresponding increase in
analytic and machine learning into the pricing strategies the the expected revenue. Both approaches leverage heatmaps
concepts of traditional revenue management for airline have and regression techniques for accurate key point detection.
evolved significantly, as it has become easier for them to
predict demand and set prices for products and services. As seen in previous sections, the effects of dynamic
This shift is most profound at a time when consumers rely pricing have also been analysed from the ethical perspective
highly on digital data in their buying behaviours needing in the literature. Stigler (2019) conducted a study with the
more accurate approach to pricing. opinion that enhanced dynamic pricing models can be a
mechanism for engaging in related abuses of price
To this end, Machine learning algorithms like the discrimination. Dynamic pricing as per the author can help
Random Forest and Gradient Boosting Machines have been improve the revenue of the airline and at the same time,
researched to determine their applicability in predicting create disparity making consumer feel that the model is
demand and tuning of the right price levels across industries tuned according to them. This has raised importance of
including the airline industry. A paper by Bertsimas and checking profits while holding the balance between business
Shioda (2018) [2] shows that these algorithms can be used to gains via fare pricing and ethical issues.
better forecast based on historical booking data. Mainly, in
line with this context, the authors put forward that through Therefore, it is clear from the above literature that the
use of big data, airlines would be able to accurately capture dynamic pricing in airline industry is a very complex field
consumer behaviour patterns in order to come up with better that incorporates technological factors technological factors,
strategic pricing decisions that will eventually improve the consumers’ characteristics, external environment factors as
overall revenue of the business. Employing these algorithms well as ethical factors. Given the strengthen usage of
within the use dynamic pricing models also improves the machine learning and big data analytics in revenue
prediction while at the same time updating the predictions as management by airlines, the future research will be required
the market moves. to tackle dynamics of the field.

Other factors concerning dynamic pricing like II. METHODOLOGY

behavioural factors, economic indicators and weather
conditions are also the focus of the recent studies. For The model utilizes the modern technologies like
example, Zhang et al. (2019) [3] make a case for the use of FBProphet, Random Forest, Streamlit, Pickle for Dynamic
macroeconomic variables including GDP growth, inflation, Price Prediction of Airlines.
and unemployment to the pricing models. The authors
identified that such factors define consumption patterns and  Data Collection
should be incorporated into effective dynamic pricing The first step is to gather data that contains features
models. Again, weather factors were adopted here to depict that have bearing on the determination of the price of airline
the effect on the customer demand for travel; therefore, tickets. The data is web scrapped and then stored in MySQL
incorporating such variables can improve the strengths of Database. The data stored is then converted into a csv file
the pricing variables. which is processed airline data file including the flight
information, customer characteristics, economic factors, the
In addition, the improvement of big data technologies rate of sales, weather, and booking habits. The first dataset is
has made it easier to gather and process massive information imported through the data loading function of Pandas.
concerning dynamic pricing. Li et al (2020) [5] noted that read_csv(), DataFrame and 10000 entries is taken to reduce
through big data analytics, airlines can get real-time computation times. Such selection ensures that the data that
information about the data e.g. from the social media, is being processed is manageable and on the same note, the
competitors’ fare and customers’ feedback. Hence, the sample obtained is strong and credible.
presented data integration approach helps to improve the
regularity of airline pricing adjustments and their  Feature Identification
anticipatory reactions to varied market environments. The In this study, the following global categorical variables
authors claim that the integration of big data technologies are defined; Flight_Number, Departure_Airport,
and machine learning forms a strong foundation for Arrival_Airport, Travel_Class, Booking_Channel. Also, the
implementing elaborate dynamic pricing solutions. list of the pricing features’ catalogue is developed which can
be categorical and numerical. E.g. economic parameters
Consumer reaction to real-time pricing is another (GDP growth rate, inflation rate), climatic factors
important a research focus. This view agrees with other (temperature, wind speed), and booking characteristics (lead
studies such as by Elmaghraby and Keskinocak (2003) [4] time). Thus, the company guarantees the consideration of all
which show that price sensitivity can be high or low necessary factors in the development of the model.
depending on some user characteristics and context. These
behavioural patterns are important for airlines willing to  Data Preprocessing
practice dynamic pricing effectively and efficiently. The Data preprocessing is less time consuming and very
study also has proven that while applying personnel pricing important in order to prepare the dataset for training. This

IJISRT25JAN1558 www.ijisrt.com 2901

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14944858
includes the process of converting date columns into  Model Serialization
numeric suitable for analysis. A custom function called After both models have been trained, they are saved for
preprocess_data is designed to execute this conversion. The later use using the pickle module of Python. This entails a
function determines the difference between the specific process of storing all the trained models as well as any other
dates for flying and booking in days and convert the flight data (like the features’ order) into files that are easily
times into minutes. In addition, it handles holiday dates in a understandable for a subsequent loading session without
similar manner. In this step, it is possible to obtain new requiring a reinforcement training. Serialization promotes
numeric columns that improve the characteristics of the machine learning applications’ effectiveness since pre-
dataset. trained models may be promptly accessed during use or
additional analysis without requiring additional
 Feature Encoding computation.
To manage categorical features in the pricing model the
process of One-Hot Encoding is used. This process is used  UI Development
to transform categorical variables into formats, which a For the purpose of interacting with the dynamic pricing
machine learning system is able to understand as binary. To model, Streamlit is used to create the front end. By using
encode these types of features, and to ensure that handling sliders and selection boxes gadgets, users can input different
of unknown categories is handled appropriately, the types of parameters like flight details, customer details,
OneHotEncoder from sklearn.preprocessing module is used. economic conditions, weather information etc. After the
Each of the encoded features is blended with the numerical collection of user inputs, scenario analysis involved comes
facts which in turn produces a merged feature set to be used with input data in format as required by the trained models.
in training the model. To achieve this, features are sampled This works based on user created outcomes to forecast and
or enumerated in a training dataset so that computer reads all hence, gives out the best pricing details propelled by certain
of them correctly. circumstances.

 Dataset Splitting III. MODEL ARCHITECTURE

After preprocessing and encoding, the dataset is
divided into training and testing sets to evaluate model The backbone of model is FBProphet and Random
performance accurately. The ‘train_test_split’ function Forest Regressor.
from ‘sklearn.model_selection’ is utilized to split the data
into 80% training data and 20% testing data. This division Random forest is an ensemble learning method builds a
allows for robust training of the model while reserving a number of decision trees when training and makes a
portion of the data for validation purposes, which helps classification by the mode of the predicted classes or a
prevent overfitting and ensures that the model generalizes regression by the mean of the predicted values. This comes
well to unseen data. in handy especially for large and high dimensional datasets
that are encountered while conducting the airline pricing.
 Model Training The model includes a number of predictor variables from the
Now that we have the training data, the next operation continuous and nominal scale including flight details,
is to train a model of the analysis of machine learning to be economic conditions, and customers aspect.
able to predict the ticket prices. Thus, in order to achieve the
best results a Random Forest Regressor is chosen because it Data preprocessing comes before the training process
is suitable for dealing with the large high dimensional with many features of categorical data being encoded using
dataset. This means that the model is fitted using the fit One-Hot Encoding, which is needed to the machine learning
method on the training dataset. This step ends up with the algorithms. The ‘RandomForestRegressor’ from the
development of the price predicting model that is able to ‘sklearn.ensemble’ library is used, and it fits the model on
make a prediction concerning a given feature. Furthermore, the training dataset. This makes Random Forest a valuable
a time series forecasting model with Facebook’s FBProphet strategy because it averages the outcomes from many trees,
library is also built on the historical ticket prices to learn the thereby reduces overfitting that is frequent with single trees
trend and seasonality which are all the time-series and improves generalization of results to new data.
components. Predicted ticket prices will be indicated at the end of the
final model in addition to input features; airlines will be in a
position to make an automatic change on ticket prices in
relation market environment occurrences.

IJISRT25JAN1558 www.ijisrt.com 2902

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14944858

Fig 1 Working of Random Forest

FBProphet is a forecasting model created by Facebook, moments, FBProphet offers information on subsequent price
which is perfect for analyzing time series data. Prophet is growth trends using historical data. This two-fold allows for
favorable when it comes to the seasonality and trend competition through the aircraft ticket pricing while at the
behavior of historical data thus making it suitable to predict same time employing historical as well as seasonal analysis
ticket prices of airlines. The model works on a simple of the demand and supply planes.
principle: This breaks down time series and factors them
into trends, seasonality, and holiday impacts, giving In order to allow users to interact with these models, a
practitioners a clear structure in which to analyse underlying web application developed using Streamlit has provided
patterns. users mechanisms for inputing different factors such as
flight information, customer profile, weather factors, and
As with the previous implementation, historical ticket economic factors. Once inputs are received, the format is
prices are also pre-adjusted before training a model in a again the same as training data, confirming that categorical
DataFrame format where the head ‘date’ is renamed to ds predicators are dummified and all the features are present.
(date stamp) and the head ‘price’ renamed to ‘y’. Seasonality The developed application then uses both of the above
is a key attribute that is easily handled through Prophet and models to come up with the best optimum price which is
Prophet can adapt and include holidays or special events that best suited for certain conditions set by the user.
might affect demands. The model is fitted with the use of the
fit() method that estimates parameters that have been made IV. RESULT
based on historical data. It is clear that when trained,
Prophet can estimate future prices while also providing an The dynamic airline pricing model's output division
expected seasonality and trend component. consists of functions which allow users to interact with the
system and displays prediction outcomes and develops
The combination of Random Forest and FBProphet visual representations. Users can learn about the model
leads to the development of a more stable dynamic pricing interaction through this section together with access to
system. Unlike Random Forest, which can work with retrieved information after engaging with the application.
different impact indicators causing actions in different

IJISRT25JAN1558 www.ijisrt.com 2903

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14944858
 User Interface Outputs: - entering relevant input parameters. Users encounter an
The application features a user interface based on orderly interface after they start the application where
Streamlit that enables users to run their pricing scenarios by multiple feature-entry fields appear.

Fig 2 User Interface

Users need to provide Flight Details through the application where multiple feature-entry fields appear. These
specification of flight number and departure along with inputs may include:
arrival airports and travel date.
Users need to provide Flight Details through the
Customer Demographics: Input fields for age, travel specification of flight number and departure along with
class (economy, business, first), and fare class. arrival airports and travel date.

Users have the ability to input present economic facts Customer Demographics: Input fields for age, travel
including GDP growth rate together with inflation rate. class (economy, business, first), and fare class.

The system allows users to input weather elements Users have the ability to input present economic facts
through separate fields which include temperature data and including GDP growth rate together with inflation rate.
wind speed measurements and precipitation measurements. The system allows users to input weather elements
through separate fields which include temperature data and
Users who submit information through the input fields wind speed measurements and precipitation measurements.
to the application witness the system generate predictions
from the collected data. The application features a user Users who submit information through the input fields
interface based on Streamlit that enables users to run their to the application witness the system generate predictions
pricing scenarios by entering relevant input parameters. from the collected data.
Users encounter an orderly interface after they start the

IJISRT25JAN1558 www.ijisrt.com 2904

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14944858

Fig 3 Scenario Selection

 Model Predictions  Time Series Forecasting:

After processing the user inputs, the application The FBProphet model generates forecasts for future
utilizes both the Random Forest model and FBProphet ticket prices based on historical data trends. This output
model to produce pricing predictions. The outputs from includes:
these models include:
 Forecasted Price Trends:
 Predicted Ticket Price: A graphical representation showing how ticket prices
The primary output is the predicted ticket price based are expected to change over time.
on the input parameters. The Random Forest model provides
a price estimate that reflects real-time market conditions and  Seasonality Effects:
influences. Insights into seasonal variations in pricing that may
affect demand during specific periods.

Fig 4 Forecasted Prices

IJISRT25JAN1558 www.ijisrt.com 2905

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14944858
 Confidence Intervals: libraries like Plotly and Matplotlib. These visual outputs
For both models, confidence intervals may be provided may encompass:
alongside predictions to indicate the reliability of the
forecasts. This information helps users understand potential The application uses Price Trend Graphs which show a
variability in ticket pricing. history of ticket prices and their predicted future values
through line charts. Users can view the price fluctuations
 Visualizations from the past as well as projected price movements through
A user-friendly interface emerges from the application this visualization format.
because it utilizes visual output elements made through

Fig 5 Showing Visualization for Booking Demands over Time

The Random Forest model shows its most impactful [2]. Elmaghraby, W., & Keskinocak, P. (2003). Dynamic
features on ticket pricing through bar chart visuals called Pricing: A Review of the Literature. Operations
Feature Importance Charts. Users learn about the main Research, 51(1), 1-20. DOI:10.1287/opre.51.1.1
elements that affect price shifts through this generated [3]. Li, Y., Wu, J., & Zhang, J. (2020). Big Data Analytics
information. for Dynamic Pricing: A Review. Journal of Business
Research, 112, 212-
The FBProphet model can depict seasonal ticket price 223. DOI:10.1016/j.jbusres.2019.10.028
patterns through plots that display effects during daily, [4]. Stigler, G.J. (2019). The Ethics of Dynamic Pricing:
weekly or monthly time spans. The visual representations Implications for Airlines. Journal of Business Ethics,
show specific intervals when demand reaches its peak or 155(3), 695-707. DOI:10.1007/s10551-017-3483-0
falls according to past records. [5]. Zhang, Y., Chen, J., & Wang, L. (2019). The Impact
of Economic Indicators on Airline Pricing Strategies:
V. CONCLUSION An Empirical Analysis. Transportation Research
Part E: Logistics and Transportation Review, 129, 1-
The output section through advanced machine learning 16. DOI:10.1016/j.tre.2019.06.001
techniques provides users with actionable insights in this [6]. Kimes, S.E., & Wirtz, J.G.A.E (2003). Perceived
dynamic airline pricing model design. The application Fairness of Demand-Based Pricing for Services: An
allows airlines to enhance their pricing effectiveness by Exploratory Study in the Airline Industry". Journal of
combining intuitive interfaces and predictive modeling with Service Research, 5(3), 235-246.
clear display features. The produced outputs enable users to [7]. Avci, T., & Kucukusta, D.(2020). Dynamic Pricing in
make distinct pricing choices in the present as well as Airline Industry with Machine Learning Techniques:
develop long-term business strategies through forecasted A Case Study on Turkish Airlines". Journal of Air
market trends and documented customer behavior patterns. Transport Management, 84, 101773.
[8]. Nair, H.S., & Raghunathan, R.(2004). Optimal
REFERENCES Dynamic Pricing Strategies for a Service Firm with
Customer Choice Behavior". Operations Research,
[1]. Chen, J., Zhang, Y., & Wang, L. (2016). Dynamic 52(2), 205–219.
Pricing Strategies for Airlines: A Review. Journal of [9]. Choi, T.M., & Cheng, T.C.E.(2019). Dynamic Pricing
Revenue and Pricing Management, 15(2), 83- for Fashion Products with Supply Chain
102. DOI:10.1057/s41272-016-0003-6

IJISRT25JAN1558 www.ijisrt.com 2906

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14944858
Considerations". International Journal of Production
Economics, 210, 30–42.
[10]. Gallego, G., & van Ryzin, G.J.(1994). Optimal
Dynamic Pricing of Inventories with Stochastic
Demand Over Finite Horizons". Management
Science, 40(8), 999–1020.
[11]. Feng, Y., & Zhao, X.(2020). Pricing Strategy in E-
commerce Based on Big Data Analysis". Journal of
Retailing and Consumer Services, 54, 102036.
[12]. Chen, Z., & Zhang, Y.(2021). Machine Learning in
Revenue Management and Dynamic
Pricing". Journal of Revenue and Pricing
Management, 20(5), 509-525.
[13]. Huang, Z., et al.(2022). A Survey on Machine
Learning Approaches for Dynamic Pricing in E-
commerce". IEEE Transactions on Systems Man and
Cybernetics: Systems, PP(99), 1-14.

IJISRT25JAN1558 www.ijisrt.com 2907