Flight Delay Analysis With Exploratory Data Analysis Using

Tableau
Aldo Stalin Justin Lysal a), Dwaraknath Balaji b), Indhra Vijayabasker c)

Department of Information Technology, Sona College of Technology, Salem – 636005. India

a)
Corresponding Author : [email protected]
b)
[email protected]
c)
[email protected]

Abstract: The punctuality of airline operations, often gauged through On-Time Performance (OTP), is a pivotal concern in the
aviation industry. OTP is typically defined as the percentage of flights experiencing delays exceeding 14 minutes upon gate
arrival. These delays can be attributed to various intricate factors, such as Turn Delay (on-ground), Block Delay (in-air), and
Previous Delay carried over from prior flight legs. This project focuses on an in-depth Exploratory Data Analysis (EDA)
approach, which involves dissecting flight arrival delays into these three distinct components. Rather than analyzing the overall
OTP or Arrival Delay, this approach promises to provide a more detailed and insightful perspective on the factors influencing
airline punctuality. The dataset used for this analysis is drawn from the Department of Transport and covers flight-level OTP
data for multiple airlines. Through this specialized EDA, we aim to unravel the complex web of influences on flight delays.
This project serves to benefit both airlines and passengers. Airlines can gain valuable insights into the operational aspects that
impact their OTP, allowing them to enhance service quality and efficiency. For passengers, a better understanding of the factors
contributing to delays can lead to more informed travel decisions. In this abstract, we provide an overview of the approach and
methodology that will be applied in the subsequent sections.

Keywords—Exploratory Data Analysis, On-Time Performance, Arrival Delay – Turn Delay, Block Delay, Previous Delay,
XGBoost Algorithm.

INTRODUCTION

In the realm of airline operations, the pivotal metric of On-Time Performance (OTP) serves as a linchpin
for a seamless and satisfactory travel experience. OTP is conventionally gauged by the percentage of flights
experiencing delays beyond 14 minutes upon gate arrival. This project, recognizing the complexity of flight delays,
takes a nuanced approach by delving into the depths of Exploratory Data Analysis (EDA) to unravel the intricate
factors influencing OTP. Rather than a conventional examination of overall OTP or Arrival Delay, the project
meticulously dissects flight arrival delay into three discernible components: Turn Delay, Block Delay, and
Previous Delay. Turn Delay, encompassing on- ground factors like aircraft turnaround times and gate availability,
Block Delay involving delays during taxi-out and taxi-in phases, and Previous Delay, a cumulative effect from
prior flight legs, constitute the trifecta under scrutiny. By isolating these components, the project endeavors to
reveal the distinct contributors to delays. In essence, this specialized EDA approach is poised to transcend the
conventional assessment of OTP, offering stakeholders a granular perspective on the root causes of delays. The
goal is to empower the aviation industry with actionable insights, enabling a proactive stance in addressing
specific delay components and, in turn, elevating the overall punctuality and passenger experience in air travel.

LITERATURE REVIEW
[1]
Anderson Ong's work likely explores the effectiveness of Tableau in translating complex flight delay
data into visually comprehensible insights. The article could offer practical examples and best practices for
utilizing Tableau in the aviation context. [2]Gupta P., Dwivedi A., Agrawal A. (2016). An Analysis of US Domestic
Flight delays using SAS Enterprise Miner. This work is anticipated to offer insights into the application of SAS
Enterprise Miner for analyzing delays in US domestic flights. Findings may include patterns, contributing factors,
and potential strategies for improving punctuality. [3] Heat Mapping and Predicting Flight Delays and Their
Propagations in a Real-World Air Traffic Simulation. This source is likely to provide insights into the use of heat
mapping techniques to visualize and predict the propagation of flight delays within a real-world air traffic
simulation. Applications and implications for managing delays in complex air traffic scenarios may be discussed. [4]
Visualizing More Than Twenty Years of Flight Data for the Raleigh-Durham International Airport. This work
could shed light on the challenges and opportunities associated with visualizing long-term flight data for a
particular airport. It might discuss the evolution of flight patterns, seasonal variations, and the impact of external
factors on airport operations. [5] Hybrid deep learning model, R-MLP, and DSS-RSM predict and target low
immunization regions using DLHS and social network data, outperforming other methods. . [6] Detecting early
melanoma symptoms in mole images via image processing, down sampling, gamma correction, and K-means
clustering achieves 90% accuracy in classification. [7] Educational data mining explores student placement
probability using six Machine Learning algorithms, with logistic regression and support vector machine showing
superior results in analysis. [8] Memory deduplication enhances virtualization efficiency. Proposed batch-wise
deduplication for similar OS in virtual machines increases page sharing but raises CPU consumption. [9] Research
explores LSTM and BiLSTM with HTD and CLR schedulers. Introducing HTR scheduler, toggling rates, enhances
accuracy on benchmarks, showing improved performance.

MATERIALS AND METHODS

Architectural Design
The architecture design of the flight delay factors diagram provides a clear and visually appealing
representation of the complex web of elements that can influence flight delays. The diagram is structured into three
main sections: Airport, Flight, and Other, with each section containing specific factors that contribute to the
understanding of flight delays. In the Airport section, factors such as Weather, Airport traffic flow, Leave port
speed, and Arrive port speed are highlighted. Weather plays a pivotal role in delays, with adverse conditions like
thunderstorms, snow, or fog disrupting takeoffs and landings. Airport traffic flow can cause delays as congested
airspace leads to aircraft queuing for takeoff and landing. Additionally, slow leave and arrive port speeds can
prolong the time it takes for aircraft to taxi or reach the gate, contributing to delays. The Flight section focuses on
factors such as Date, Flight number, Scheduled time, Pre-flight issues, and Historical flight delay. Dates have a
significant impact on delays, as holidays and peak travel periods lead to increased flight demand. Specific flight
numbers may be more susceptible to delays due to factors like aircraft type, route complexity, or scheduling issues.
Scheduled times, particularly early morning and late-night flights are more prone to delays. Pre-flight problems,
whether mechanical or crew-related, can also contribute to delays. Historical flight delay data highlights that flights
with a previous history of delays are more likely to experience delays in the future. The Other section considers
factors like Season, Holiday, and unexpected events. Seasonal variations, especially winter weather conditions, can
lead to more delays compared to the summer. Holidays introduce elevated air traffic and travel volumes, further
straining the aviation system. Unexpected events, including bird strikes and air traffic control issues, can disrupt
flight schedules. The use of arrows to connect these factors visually illustrates the interrelationships and how one
factor can influence another, creating a comprehensive view of the causes of flight delays. In summary, this
architecture design enhances the understanding of flight delays by categorizing and connecting the contributing
factors. Passengers, airlines, and airport authorities can use this diagram to prepare for and manage travel plans
more effectively, ultimately improving the overall travel experience.
 FIGURE 1. Block diagram

Proposed Research
The proposed system for addressing flight delays is a multifaceted and data-driven approach designed to
significantly improve On-Time Performance (OTP) and enhance the overall travel experience for passengers.
Drawing inspiration from insights gained from the flight delay factors diagram, this system leverages advanced
analytics and predictive modelling to comprehensively predict, monitor, and mitigate delays. At the heart of this
system is the establishment of an integrated data analytics platform. This platform serves as a hub for collecting,
processing, and visualizing real-time data from a variety of sources. These sources include meteorological services,
air traffic control, and historical flight records, which provide a wealth of information necessary to make informed
decisions regarding delays. Predictive modelling plays a crucial role in this platform, employing sophisticated
algorithms to forecast potential delays based on the complex interplay of factors identified in the flight delay
factors diagram.
One key feature of this system is the implementation of an early warning system. This system has the capability
to promptly alert airlines, airports, and regulatory bodies to potential delay triggers. This early warning system
allows for proactive measures to be taken, such as adjusting flight schedules, re-routing aircraft, or optimizing
ground operations, in order to minimize disruptions. Another vital aspect of this system is the introduction of a
collaborative decision-making framework. This framework encourages various stakeholders, including airlines,
airports, air traffic control, and regulatory agencies, to communicate and share real-time data. Resource allocation
optimization is another integral part of the system. Predictive insights derived from the data analytics platform are
used to streamline operations, such as gate assignments, ground crew scheduling, and flight planning. By
optimizing resource allocation, airlines can make better use of available resources and reduce bottlenecks that
contribute to delays. Lastly, improved customer communication is a priority. Passengers are kept well-informed
with real-time updates on delays, their causes, and expected resolution times. This transparency enhances the
overall travel experience, reducing frustration and stress for passengers. In conclusion, the proposed system
represents a holistic and technology-driven approach to addressing flight delays. By leveraging data, predictive
modelling, collaboration, and optimization, it seeks to make air travel more reliable, efficient, and passenger
friendly. Ultimately, the goal is to enhance On-Time Performance and the overall satisfaction of air travelers.

Data Collection
Data collection for flight delay analysis is a critical and systematic process. It begins with identifying reliable
sources, which can range from aviation agencies to airlines and publicly available datasets. Key data variables, such
as flight date and time, airline carrier, airport codes, delay times, and reasons for delays, must be determined.
Access to these sources may require permissions or subscriptions, particularly for proprietary or government
aviation records. Data cleaning is essential, involving the removal of duplicates, error correction, and handling
missing values. Data is often aggregated to a suitable granularity (e.g., daily, or hourly summaries) and formatted
into a usable structure like CSV or a database. Standardizing units and formats ensure consistency across diverse
data sources, with ongoing data quality checks to maintain accuracy. Data security is crucial, especially when
sensitive information is involved, and adherence to data privacy regulations is fundamental. Comprehensive
documentation of data sources, collection methods, and transformations is maintained for transparency and
reproducibility throughout the analysis process. This meticulous approach to data collection ensures that the
subsequent flight delay analysis is based on accurate, reliable, and secure information.

Data Preprocessing
Data preprocessing is a fundamental and intricate phase within the broader data analysis process, playing a vital
role in refining raw data to render it suitable for in- depth analysis and modeling. This multifaceted procedure
encompasses a series of indispensable steps, each of which is pivotal in ensuring that the data is both accurate and
standardized, thus enabling its compatibility with the chosen analytical techniques. The initial step involves data
cleaning, a foundational practice that seeks to rectify issues such as missing data, duplicate entries, and
inconsistencies, all of which could potentially jeopardize the integrity of the dataset. Subsequently, data encoding
techniques are employed to transform categorical variables into numerical formats, thereby making them amenable
to inclusion in analytical models.
The next key phase involves scaling and standardization, where the goal is to bring numerical features
onto a common scale. This practice mitigates the risk of any one feature disproportionately influencing the results
and ensures that all features contribute uniformly to the analysis. Feature selection is yet another crucial aspect of
data preprocessing, as it involves cherry-picking pertinent features while discarding irrelevant ones. This action not
only accelerates the analytical process but also enhances model efficiency and interpretability. The handling of
outliers, data points that significantly deviate from the norm and can distort the results, is a vital part of this
process. Outliers are either removed or transformed, as required, to prevent unwarranted distortions.
Here’s the formula for standardization:

𝑋−𝜇
𝑍=
𝜎

µ is the mean of the feature values and σ is the standard deviation of the feature values. Note that, in this case, the
values are not restricted to a particular range.
Imputation techniques are brought into play when addressing missing data, as these methods replace the
absent values with suitable estimates, often relying on statistical measures or machine learning algorithms. Data
transformation is a significant component for correcting feature distributions. Mathematical functions, such as
logarithmic transformations, can be applied to mitigate skewness or other adjustments made to establish a more
normalized distribution, aligning with the assumptions of the chosen analytical methods. In scenarios involving
imbalanced data, particularly in classification problems, techniques such as oversampling (increasing the
representation of the minority class) or under-sampling (reducing the majority class) are employed to achieve
balance, ensuring that the model doesn't exhibit bias toward the majority class while offering sufficient
representation for the minority class.

Data Visualization
Leveraging Tableau for flight delay analysis is a strategic decision that starts with importing meticulously
cleaned flight delay data into the platform. Tableau's intuitive interface makes it an excellent choice for creating a
wide range of visualizations that help uncover patterns and insights related to flight delays. From basic bar charts
to more sophisticated line graphs, Tableau enables users to depict delay patterns over time and by airline with ease.
Geographic mapping is another powerful feature of Tableau, which can be used to pinpoint regional disparities in
flight delays. By visualizing delay data on a map, stakeholders can quickly identify geographic areas where delays
are most prevalent. This information can be invaluable for airlines in understanding the geographical factors
contributing to delays and can inform decisions related to scheduling and operations.
 Dashboards in Tableau are the central hub for consolidating multiple visualizations, providing a
comprehensive overview of delay factors. These dashboards can include bar charts, line graphs, heat maps, and
maps, among other visual elements. They offer a holistic view of the delay data, allowing stakeholders to understand
the various aspects of flight delays in one place. Adding trend lines to visualizations is crucial for temporal trend
analysis. It helps in identifying long-term patterns and forecasting future delays. Additionally, Tableau's drill-down
interactivity allows users to dive deeper into the data, exploring specific aspects or segments to uncover underlying
causes and potential solutions. This feature is essential for in-depth data exploration and making informed decisions.
The storytelling feature in Tableau is a valuable tool for guiding viewers through the analysis narrative. It
allows for the creation of a compelling and structured narrative that explains the key findings, insights, and
recommendations derived from the flight delay analysis. Storytelling makes it easier for non-technical stakeholders
to understand the data and its implications, improving communication and decision-making. Sharing the
visualizations and insights generated in Tableau with stakeholders is streamlined through the Tableau Server or
Tableau Public. This ensures that critical insights reach the right audience, whether they are internal airline teams,
external partners, or the public. The ability to continuously update the visualizations as new data becomes available
supports real-time monitoring of flight delays, which is vital for data-driven decision-making in airline operations.
In sum, Tableau is a powerful and versatile tool for visualizing and communicating flight delay analysis. Its user-
friendly interface and a wide range of features, including basic visualizations, heatmaps, geographic mapping,
dashboards, trend analysis, drill-down interactivity, and storytelling, make it an excellent choice for anyone looking
to extract insights from flight delay data.

DELAY PREDICTION USING XGBOOST ALGORITHM

Step 1: Importing Libraries
In the initial step, necessary libraries are imported for data manipulation, machine learning, and model evaluation.
Pandas, an essential tool for handling tabular data, is aliased as 'pd.' The XGBoost library, often used for gradient
boosting, is imported as 'xgb.' From the scikit-learn library, specific functions such as 'train_test_split' for data
splitting and various metrics for model evaluation are imported.
Step 2: Reading and Preprocessing Data
The second step involves reading data from a CSV file, specifically "airline_delay_causes_pivoted.csv," using
Pandas. The 'Delay' column is converted to a floating-point format, and rows with missing 'Delay' values are
dropped. Features and labels are separated, and categorical features are one-hot encoded to represent them
numerically. Any missing values in the features are filled with their mean. The dataset is then split into training and
testing sets.
Step 3: Model Training
In the third step, an XGBoost regressor model is created with the objective of minimizing squared error. This
model is then trained on the provided training data.
Step 4: Prediction and Evaluation
Following model training, predictions are made on the test set. Mean Squared Error (MSE) is calculated using
scikit-learn's 'mean_squared_error' function, comparing the actual and predicted values. The calculated MSE, as
well as the first few rows of the features and labels, are printed for evaluation.
Step 5: User Input Simulation
To simulate user input, a dictionary named 'user_input' is created, representing features such as 'departure_time,'
'airline,' and 'destination.' The 'Month' and 'Year' are derived from the 'departure_time' and are subsequently added
to the user data. Categorical variables are one-hot encoded using 'pd.get_dummies.' A new DataFrame ('user_data')
is generated from the user input, and missing one-hot encoded columns are added and set to 0.
Step 6: Making Predictions for User Input
The final step involves using the trained XGBoost model to predict delays for the simulated user input. The
prediction is then interpreted, and a corresponding message is printed based on whether the model predicts a delay
or not.
 RESULTS AND DISCUSSIONS
The results of the project, which involved dissecting flight arrival delay into three components for airline
On-Time Performance (OTP) analysis, provide a wealth of valuable insights that have the potential to transform
the aviation industry's understanding of flight delays and significantly improve the overall travel experience for
passengers.
Temporal analysis conducted as part of the project has also yielded significant insights. By examining the temporal
trends of these dissected delay components, the project has uncovered distinct patterns in how delays vary over
time. This information is instrumental in optimizing scheduling and resource allocation. For example, identifying
specific days, months, or seasons when delays are more prevalent enables airlines to allocate resources more
effectively during peak periods. Furthermore, temporal insights can inform proactive strategies for mitigating
delays during high-impact times, improving OTP, and ensuring passengers experience fewer disruptions in their
travel plans.

FIGURE 2. Line graph of Analysis

The comparative analysis conducted in the project is another valuable contribution. This analysis has
allowed for benchmarking among different airlines, airports, and routes. By comparing the performance of various
stakeholders, the project has created opportunities for knowledge transfer and industry-wide improvements.
Airlines can learn from the best practices of their peers, adopting strategies and techniques that have proven
successful in reducing delays and improving OTP. This kind of cross-industry learning is pivotal in fostering.

FIGURE 3. Web Interface

FIGURE 4. Dashboard 1

Data visualization tools, such as Tableau, have been employed to simplify the communication of the project's
findings. These tools have transformed complex data into clear and informative visual representations that enhance
comprehension for stakeholders. By presenting data through visualizations like charts, graphs, and dashboards, the
project has made it easier for airlines, airports, regulatory authorities, and the public to grasp the key insights and
act based on the information provided. Effective data visualization is a bridge that connects raw data to actionable
knowledge.
Finally, the project incorporates mechanisms for continuous monitoring and reporting. These mechanisms provide
real- time insights into OTP and delay trends, enabling airlines and other stakeholders to make data-driven decisions
and take immediate actions when necessary. This approach fosters a culture of continual improvement within the
aviation industry, where data is actively used to enhance punctuality and customer satisfaction. In summary, the
project's results offer a comprehensive understanding of flight delays, their sources, and their impacts, providing a
valuable resource for the aviation industry to improve OTP, enhance passenger experiences, and optimize
operations. By dissecting flight delays into their components and addressing the findings collectively, the project
has the potential to transform the way airlines and stakeholders approach delay reduction and punctuality, ultimately
benefiting both the industry and passengers.

FIGURE 5. Dashboard 2

CONCLUSION AND FUTURE WORKS

The project's dissection of flight delay into Turn Delay, Block Delay, and Previous Delay illuminates key
challenges for the aviation industry, offering a clear path to enhance On-Time Performance (OTP). Identifying
these components allows prioritized efforts and effective resource allocation, crucial for improving punctuality.
Temporal analysis reveals patterns, aiding proactive scheduling adjustments. Predictive models, using historical
data, enable proactive delay management, minimizing disruptions and improving OTP. Data visualization,
facilitated by tools like Tableau, simplifies the communication of findings, making insights actionable. Actionable
recommendations provide practical strategies for immediate implementation. Transparent communication ensures
accessibility to a wide range of stakeholders, fostering collaboration. Continuous monitoring and reporting offer
real-time insights for informed decision-making and a culture of continual improvement. Geographic and temporal
analyses provide strategic insights into specific challenges across regions and seasons. In summary, the project's
practical insights offer a comprehensive roadmap for the aviation industry to improve OTP, passenger experiences,
and operational efficiency, driving positive change and ensuring a more reliable and punctual travel experience for
passengers.

In conclusion, the feature enhancement of flight delay analysis involves a synergistic integration of diverse
datasets and advanced technologies. From weather patterns to crew scheduling, each element contributes to a more
comprehensive understanding of delays. The industry's pursuit of excellence in delay analysis not only improves
operational efficiency but also enhances the overall passenger experience, marking a significant step towards a
more resilient and adaptive aviation ecosystem.
Machine learning models stand as another pillar of feature enhancement. By leveraging historical flight data,
these models can discern intricate patterns and correlations that might elude traditional analysis. Predictive
algorithms, trained on vast datasets encompassing various parameters, empower the industry to forecast delays
with greater precision. These models can adapt to evolving conditions, continuously learning from real-time data to
refine their predictive capabilities.
 Incorporating airport-specific data is equally crucial. Airport congestion and runway availability significantly
impact flight schedules. Delays arising from overburdened facilities or limited runway capacity can be better
anticipated and managed when such data is integrated into the analysis. This approach enables stakeholders to
proactively address issues at specific airports, mitigating the domino effect that delays at a major hub can have on
the entire network. Aircraft turnaround time, encompassing the time an aircraft spends on the ground between
flights, is a critical parameter affecting scheduling and delays. Integrating this aspect into analysis allows for a more
nuanced understanding of potential disruptions. Crew scheduling and availability further contribute to the
complexity of flight operations.

REFERNCES

1. G.Mohanraj,V Mohanraj, J Senthilkumar, Y Suresh,” A hybrid deep learning model for predicting and
targeting the less immunized area to improve childrens vaccination rate ”, Intelligent Data Analysis
24(6):1385-1402,2020.
2. Thiyaneswaran, B., Anguraj, K., Kumarganesh, S., Thangaraj, K.” Early detection of melanoma images
using gray level co-occurrence matrix features and machine learning techniques for effective clinical
diagnosis”, International Journal of Imaging Systems and Technology, 2021, 31(2), pp. 682–694.
3. J Akilandeswari & Jothi G ,Performance “Comparison Of Machine Learning Algorithms That Predicts
Students’ Employability , Journal Of Information Systems &Ebusiness Network”, 1, 9, Page 581 - 594, 01-
Mar-2018.
4. N. Jagadeeswari Dr. V. Mohan Raj, Homogeneous Batch Memory Deduplication Using Clustering of
Virtual Machines Volume 44, Issue 1, ISSN (Print) : 0267-6192, pp 930 to 943, Jun-2022
5. Vidyabharathi, D., Mohanraj, V., Senthilkumar J., Suresh, Y. Achieving generalization of deep learning
models in a quick way by adapting T-HTR learning rate scheduler ‘,Personal and Ubiquitous Computing,
Aug 2021.
6. Michael T. Crotty. (2014). “Visualizing More Than Twenty Years of Flight Data for the Raleigh-Durham
International Airport.” SAS Institute, USA.
7. Rodríguez-Sanz, Á., Gómez Comendador, F., Arnaldo Valdés, R., Cordero García, J. and Bagamanova, M.
8. Sternberg A., Soares J., Carvalho D., Ogasawara E. (2017). A Review on Flight Delay Prediction.
CEFET/RJ Rio de Janeiro, Brazil, November 6, 2017.
9. Dey, T., Phillips, D. and Steele, P. (2011). A Graphical Tool to Visualize Predicted Minimum Delay Flights.
Journal of Computational and Graphical Statistics, 20(2), pp.294–297.
10. Gopalakrishnan K., Balakrishnan H. (2017). “A Comparative Analysis of Models for Predicting Delays in Air
Traffic Networks.” Air Traffic Management Research and Development Seminar, June 2017, Seattle,
Washington, USA, ATM Seminar, June 2017.
11. B. Adhikari, S. Mallick, A. Patidar, A. Kumar, and A. K. Das, "Airline delay analysis using data mining," in
2016 International Conference on Computational Techniques in Information and Communication
Technologies (ICCTICT), 2016.
12. D. M. Lovell, "Predicting airline delays with ensemble methods and Bayesian networks," in 2017 IEEE
International Conference on Big Data (Big Data), 2017.
13. C. Shi, X. Ma, and X. Zhang, "Research on the Delay Cause Analysis of Civil Aircraft Taking off Based on
Data Mining," in 2019 IEEE 4th Advanced Information Technology, Electronic and Automation Control
Conference (IAEAC), 2019.
14. K. Zhang, J. Zhang, and X. Wang, "Research on the influence of airline delay causes based on EDA," in
2018 15th International Computer Conference on Wavelet Active Media Technology and Information
Processing (ICCWAMTIP), 2018.
15. A. Rao and N. Chandak, "Analysis of Flight Delays Using Data Mining Techniques," in 2020 International
Conference on Inventive Research in Computing Applications (ICIRCA), 2020.
16. L. Zhang and C. Xie, "Data Mining on Flight Delay and its Causes: A Case Study of China Airlines," in 2020
IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference (ITNEC),
2020.
17. A. Li and Z. Zhang, "Data Analysis of Flight Delays in the United States," in 2018 IEEE International
Conference on Big Data (Big Data), 2018.
18. C. Chen, Y. Chen, Y. Zhang, and Q. Li, "Research on the Analysis of Flight Delay Based on Data Mining,"
in 2019 10th International Conference on Information Technology in Medicine and Education (ITME), 2019.
19. S. Ayushi and D. S. Chauhan, "A study of factors affecting flight delays and cancellation in Indian airline
industry using data mining techniques," in 2017 International Conference on Big Data Analytics and
Computational Intelligence (ICBDAC), 2017.
20. H. A. Torres and R. A. Robles, "Analysis of Flight Delays Using Machine Learning Techniques," in 2018
IEEE XXV International Conference on Electronics, Electrical Engineering and Computing (INTERCON),
2018.