SN TITLE AUTHOR METHODOLOGY CHALLENGES/LIMITATIONS

O USED
21 Prediction of flight delay Desmond Bala The methodology Data Variability: Handling
using deep operator network Bisandu , involves an diverse and complex
with gradient-mayfly Irene Moulitsas Optimization- datasets with high
optimisation algorithm Driven Deep variability in delay patterns.
Learning Model, Computational Intensity:
where a novel Training and optimizing the
approach based on DeepONet model with
DeepONet is GMOA requires significant
extended with computational resources.
optimization Generalization: Ensuring
techniques to the model performs
improve flight effectively across different
delay prediction. geographic regions and
The Box-Cox datasets beyond the US
Transformation is Bureau of Transportation
data.
applied for data
Dependence on
conversion to
Optimization: Model
minimize error
performance heavily relies
rates. A Deep
on the efficacy of the
Residual Network
GMOA for weight selection,
is utilized for which might limit
feature fusion adaptability to alternative
before training the optimization algorithms.
model, enhancing
the data
representation.
The Gradient
Mayfly
Optimization
Algorithm
(GMOA) is
employed to
determine optimal
weights for the
DeepONet model,
improving
predictive
performance.
Flight On-Time
Data from the US
Bureau of
Transportation
Statistics is used
for model training
and validation.

22 Airline delay prediction by Khaksar, H., The methodology The parameters identified
 machine learning algorithms Sheikholeslami, involves Bayesian for delay prediction differ
A. Modeling, a between U.S. and Iranian
probabilistic airline networks, which may
approach used to limit the generalization of
estimate flight delay the model to other regions.
occurrences by The accuracy of the
modeling the proposed methods is over
uncertainty and 70%, which may not be
relationships sufficient for precise flight
between different delay management in
influencing factors. certain cases.
A Decision Tree The dataset used for the
model is employed U.S. network may not fully
to predict delays by represent other regions or
splitting data based global aviation conditions.
on decision rules
related to factors
such as weather and
departure times.
Cluster Classification
groups flights with
similar delay
patterns, helping
identify factors that
influence delays.
Random Forest, an
ensemble method,
combines multiple
decision trees to
improve accuracy. A
Hybrid Method
integrates different
machine learning
techniques, such as
decision trees and
Random Forest, to
further enhance the
prediction accuracy.
23 Chained Predictions of Flight Jun Chen The methodology Data Availability and
Delay Using Machine Learning and involves using Multi- Quality: The model's
Meng Li label Random Forest performance can be
Classification, an influenced by the quality
ensemble learning and availability of data.
approach to classify Incomplete or poor-quality
multiple delay labels data (e.g., missing
based on features departure times or arrival
such as weather, delays) can reduce
aircraft type, and prediction accuracy.
departure times. Unforeseen Delays: While
This enables the the model captures delays
prediction of flight based on historical
 delays across patterns, it may struggle
various delay types with unforeseen or extreme
(e.g., early, on-time, conditions, such as
delayed). An unexpected weather events
Optimal Feature or emergency situations
Selection Process is Generalization Across
implemented to Airports and Regions
enhance model Computational Complexity.
performance by
selecting the most
relevant features,
which reduces
computational
complexity and
ensures the focus is
on the most
impactful factors.
Additionally, the
Delay Propagation
Model is introduced
to track how delays
propagate through
the airport network,
with a Chained
Delay Prediction
Model designed to
connect initial
delays to those
along the entire
flight itinerary,
iteratively updating
predictions to
improve accuracy
24 Flight delay prediction based Maryam The methodology Despite the improvements
on deep learning Farshchian involves utilizing in model performance,
and Levenberg-Marquart Yazdi, Seyed Deep Learning (DL) limitations include the
algorithm Reza Kamel, to predict flight imbalanced data, which,
Seyyed Javad delays, leveraging although addressed using
Mahdavi its capability to undersampling, may still
Chabok ,Maryam automatically introduce biases in the
Kheirabadi extract relevant prediction process. The use
features from large, of Deep Learning
complex datasets. A techniques, especially with
Stack Denoising the addition of the
Autoencoder (SDA) denoising autoencoder and
is incorporated to optimization algorithms,
address noisy flight increases the complexity of
delay data, model training and tuning,
enhancing the necessitating substantial
model’s accuracy. computational resources.
The Levenberg- Furthermore, while the
 Marquardt (LM) model outperformed
algorithm is applied existing methods like RNN,
for optimizing the its generalization
weight and bias
values in the model.
To address the
imbalance in the
dataset, an
undersampling
method is used. The
study compares the
performance of
three models: SDA-
LM, SAE-LM, and
SDA, using
evaluation metrics
such as precision,
accuracy, sensitivity,
recall, and F-
measure.
25 Distribution Prediction of Ziming Wang, The study employed The study’s limitations
Strategic Flight Delays via Chaohao Liao, various machine include the reliance on a
Machine Learning Methods Xu Hang, learning algorithms single dataset, which may
Lishuai Li, to predict flight affect the generalization of
Daniel Delahaye, delays at the the results. Excluding ATFM
Mark Hansen strategic stage, delay data could influence
critical for airport the accuracy of the
slot allocation and predictions. The model's
airline operations. accuracy, while strong, still
Using empirical data leaves room for further
from Guangzhou improvements. The
Baiyun International confidence levels achieved
Airport, the are reasonable but may
researchers vary across different
developed three airports. Refining the model
evaluation metrics could enhance its
to assess prediction applicability in real-world
accuracy. The scenarios.
models, designed to
estimate the
distribution of
delays rather than
predicting a single
value, achieved high
accuracy in
predicting
departure and
arrival delays at 0.65
and 0.50 confidence
levels, respectively.
Notably, these
 predictions were
made without the
inclusion of Air
Traffic Flow
Management
(ATFM)

26 A deep learning approach Kaiquan Cai; Yue In this study, a flight The proposed method has
for flight delay prediction Li; Yi-Ping delay prediction limitations, including
through time-evolving graphs Fang; Yanbo Zhu approach is reliance on
developed using a comprehensive and
Graph Convolutional accurate graph-structured
Network (GCN) to data, which may not
model time-varying always be available,
spatial interactions especially during
in multi-airport emergencies. The model
networks. The may struggle to
method generalize across
incorporates a different airport
temporal networks. Its complexity
convolutional block can lead to high
based on the computational costs,
Markov property to
particularly with large-
capture the evolving
scale data. Real-time
patterns of flight
adaptation to rapidly
delays over time. To
changing conditions
address the issue of
remains challenging.
incomplete graph
inputs, an adaptive
Despite good accuracy,
graph convolutional
the method may not
block is integrated perform optimally in
to reveal hidden dynamic, real-world
spatial interactions scenarios.
between airports.
Extensive
experiments show
that this approach
outperforms
benchmark
methods, achieving
significant accuracy
improvements with
acceptable
execution time.
27 Flight delay prediction Mingdao Lu, The paper proposes The limitations of the
using gradient Peng Wei, a flight delay proposed flight delay
boosting machine Mingshu He and prediction model prediction model include its
learning classifiers Yinglei Teng that incorporates reliance on historical data,
both temporal and which may not fully capture
spatial unforeseen factors like
characteristics to extreme weather events or
 address the last-minute schedule
complexities and changes. Additionally, the
non-linear model may struggle with
relationships in real-time data integration
flight delay factors. due to the dynamic nature
The model considers of air traffic, potentially
influences such as affecting its accuracy. The
preceding flights, complexity of the model
airport conditions at might also result in longer
departure and processing times, especially
landing, and the when applied to large
overall situation of datasets, limiting its
flights on the same scalability.
route. Using
machine learning
algorithms, the
model is trained on
historical data and
tested with real-
time data to
enhance prediction
accuracy.
28 Using scalable data mining LORIS The study utilizes The study's limitations
for predicting flight delays BELCASTRO, parallel algorithms include the reliance on
FABRIZIO implemented as historical data, which may
MAROZZO, MapReduce not account for sudden,
DOMENICO programs executed unpredictable changes in
TALIA PAOLO on a Cloud platform weather or operational
TRUNFIO, to analyze and mine factors. The model
airline flight and primarily focuses on
weather weather conditions, which
observation might overlook other
datasets. The model significant factors
incorporates flight influencing delays, such as
information such as air traffic control or
the origin airport, mechanical issues.
destination airport, Additionally, the model’s
scheduled accuracy is dependent on
departure and the quality and
arrival times, along completeness of the input
with weather datasets, and it may not
conditions at both generalize well to different
airports. It predicts regions or airports with
the arrival delay different operational
based on these characteristics.
variables and
evaluates its
performance using
accuracy and recall
metrics at different
delay thresholds (15
 minutes and 60
minutes).

29 A hybrid Bashayer
artificial intelligence approach Alharbi1 and
to predict flight delay Master Prince2

30

In [21], the study by Desmond Bala Bisandu and Irene Moulitsas, titled "Prediction of Flight Delay
Using Deep Operator Network with Gradient-Mayfly Optimisation Algorithm," explores an
optimisation-driven deep learning approach for predicting flight delays the methodologies used are
DeepONet Framework, Gradient-Mayfly Optimisation Algorithm (GMOA), Box-Cox Transformation,
Deep Residual Network, Flight On-Time Data, and 4-Fold Cross-Validation. valuation Metrics: Models
were evaluated using metrics such as RMSE (0.0765), MSE (0.0058), MAE (0.0049), and MAPE
(0.0043). Accuracy: The proposed GMOA-based DeepONet achieved an impressive accuracy rate of
92.39%. The model correctly identified approximately 91% of on-time flights and achieved an 84%
classification accuracy for delayed flights.

In[22] , the study by Khaksar and Sheikholeslami, titled "Airline Delay Prediction by Machine
Learning Algorithms," explores the use of machine learning techniques for flight delay prediction.
The methodologies used are Bayesian Modeling, Decision Tree, Cluster Classification, Random
Forest, and Hybrid Method. The evaluation metrics showed that the methods achieved over 70%
accuracy in predicting delay occurrence and magnitude. However, limitations include the region-
specific parameters affecting delays, such as visibility and fleet age, which may not generalize to
other regions or global aviation conditions. Additionally, the accuracy may not be sufficient for
precise flight delay management in some cases.

In[23] the research of Jun Chen and Meng Li titled “Chained Predictions of Flight Delay Using
Machine Learning”emphasizes the significance of airline delay analysis and prediction, highlighting
the potential of machine learning in improving flight operations. Among the methodologies used,
Multi-label Random Forest Classification was utilized to predict various types of delays (e.g., early,
on-time, delayed) by considering multiple features such as weather, aircraft type, and departure
times. The study also implemented an optimal feature selection process to enhance model
performance by focusing on the most influential factors, thus reducing complexity. The Delay
Propagation Model and Chained Delay Prediction Model were proposed to capture and predict how
delays propagate through the airport network, with iterative updates of departure delays improving
prediction accuracy. The experiments were conducted using a test set of 26,693 flights operated in
January 2017 within the ORD-related airports network. The study demonstrated that the iteration
number, representing the number of times a flight is operated during the day, plays a crucial role in
the delay prediction process. Additionally, the study showed that 97.52% of flights had less than 6
iteration numbers in the ORD-related air traffic system, with the maximum iteration number being
10.
In[24] ,the study by Maryam Farshchian Yazdi, Seyed Reza Kamel, Seyyed Javad Mahdavi
Chabok ,Maryam Kheirabadi titled “Flight delay prediction based on deep learning and Levenberg-
Marquart algorithm”focuses on evaluating flight delay prediction using four different methods. The
methodologies employed include LightGBM, which is an efficient gradient boosting framework, for
predicting flight delays based on multiple input features. The performance of each method was
measured in terms of accuracy, Mean Squared Error (MSE), and Mean Absolute Error (MAE). The
results showed that the LightGBM model outperformed the other methods, achieving an accuracy of
0.86. Statistical analysis revealed that 89% of the delays were attributable to wind. To further
enhance the model's understanding of delay causes, the researchers established dependencies
between variables using a decision tree and association rules. Additionally, the probability of delay
occurrence was computed using Linear Regression, incorporating the dependencies discovered
earlier.’
In[25], the study by Ziming Wang,Chaohao Liao,Xu Hang,Lishuai Li,Daniel Delahaye,Mark Hansen
titled “ Distribution Prediction of Strategic Flight Delays via Machine Learning Methods”
study includes the exclusion of ATFM (Air Traffic Flow Management) delay data may lead to less
accurate predictions. While the model achieved a prediction accuracy of 0.80 for departure delays at
a 0.65 confidence level and 0.80 for arrival delays at a 0.50 confidence level, further improvements
are needed. The model's performance may vary across different airports, and its applicability could
be enhanced by incorporating more diverse data.

In[26], the study by Kaiquan Cai; Yue Li; Yi-Ping Fang; Yanbo Zhu titled ” A deep learning approach
for flight delay prediction through time-evolving graphs” the study investigates flight delay
prediction for major Chinese airports, including Beijing Capital International Airport (ZBAA), Shanghai
Pudong International Airport (ZSPD), Xian Xianyang International Airport (ZLXY), and Guangzhou
Baiyun International Airport (ZGGG). The proposed model is designed to meet both long-term and
short-term forecasting requirements. For short-term predictions, the model provides the following
results: for one hour ahead, the average prediction error is 5.884±0.072 minutes; for two hours
ahead, the average error is 6.123±0.089 minutes; and for three hours ahead, the average error is
6.252±0.093 minutes. These results highlight the model's accuracy in predicting delays for multiple
airports in China.

In [27] the study by Mingdao Lu, Peng Wei, Mingshu He and Yinglei Teng titled “Flight delay
prediction using gradient boosting machine learning classifiers” the study introduces a flight delay
prediction model using machine learning, which leverages temporal and spatial characteristics to
improve prediction accuracy. The model incorporates factors such as preceding flights, the status of
departure and landing airports, and the overall situation of flights on the same route. The model is
trained using historical data and tested with actual data. The test results indicate that the model
achieves a prediction accuracy of around 85%, with a mean absolute error (MAE) of approximately
4.5 minutes, demonstrating its effectiveness in predicting flight delays and offering a promising
method for airlines, airport resource allocation, and strategic planning in the aviation industry.

In [28] the study by LORIS BELCASTRO, FABRIZIO MAROZZO, DOMENICO TALIA PAOLO TRUNFIO
titled “Using scalable data mining for predicting flight delays” evelops a flight delay prediction model
focused on predicting arrival delays based on weather conditions. The model uses flight information,
such as origin and destination airports, scheduled departure and arrival times, along with weather
data at both airports. The system is implemented using parallel algorithms in a MapReduce
framework, executed on a cloud platform, to analyze and mine airline flights and weather
observation datasets. The model's performance is evaluated with different delay thresholds,
achieving an accuracy of 74.2% and a recall of 71.8% for delays over 15 minutes, and an accuracy of
85.8% and a recall of 86.9% for delays over 60 minutes, demonstrating its effectiveness in predicting
flight delays.
REFERENCES
[21] Bisandu, D. B., & Moulitsas, I. (2024). Prediction of flight delay using deep operator network with
gradient-mayfly optimisation algorithm. Expert Systems With Applications, 247, 123306.

[22] Khaksar, H., & Sheikholeslami, A. (2019). Airline delay prediction by machine learning
algorithms. Scientia Iranica, 26(5), 2689-2702.

[23] Chen, J., & Li, M. (2019). Chained predictions of flight delay using machine learning. In AIAA
Scitech 2019 forum (p. 1661).

[24] Yazdi, M. F., Kamel, S. R., Chabok, S. J. M., & Kheirabadi, M. (2020). Flight delay prediction
based on deep learning and Levenberg-Marquart algorithm. Journal of Big Data, 7(1), 106.

[25] Wang, Z., Liao, C., Hang, X., Li, L., Delahaye, D., & Hansen, M. (2022). Distribution prediction of
strategic flight delays via machine learning methods. Sustainability, 14(22), 15180.

[26] Cai, K., Li, Y., Fang, Y. P., & Zhu, Y. (2021). A deep learning approach for flight delay prediction
through time-evolving graphs. IEEE Transactions on Intelligent Transportation Systems, 23(8), 11397-
11407.

[27] Lu, M., Peng, W., He, M., & Teng, Y. (2021). Flight delay prediction using gradient boosting
machine learning classifiers. Journal of Quantum Computing, 3(1), 1.

[28] Belcastro, L., Marozzo, F., Talia, D., & Trunfio, P. (2016). Using scalable data mining for
predicting flight delays. ACM Transactions on Intelligent Systems and Technology (TIST), 8(1), 1-20.

[29]