Air Passenger Traffic Forecasting with Machine Learning

This presentation discusses the use of machine learning for forecasting air passenger traffic. It covers traditional forecasting approaches, project aim, execution plan, data collection and analysis plan, proposed artefact, and more.

Air Passenger Traffic
Forecasting with
Machine Learning
Student’s Name
Instructor’s Name
Course Code
Date

Table of Contents
• Introduction
• Problem Statement
• Traditional air traffic forecasting approaches
• Project Aim
• Project Execution Plan
• Approach
• Design
• Data collection/ analysis plan
• Historical data
• Analysis
• Proposed artefact
• About the artefact
• Design of the artefact models
• Implementation Approach
• Development models

Table of Contents (Contd.)
• Forecasting with intelligence
• Results
• Trends
• Responsibilities
• Roles
• Limitation
• Comparison with the current system
• Further Research
• Conclusion
• References

Introduction
 Analysis of the air traffic movements help in
forecasting the future need and demands of
service in the airport.
 Forecasting help is determining the improvements
required to meet these demands
 Analysis of the air traffic data for 20 major cities
have been analysis to propose the forecasting
model.

Problem Statement
 There are certain negative consequences along
with the substantial economic benefits, these are.
 Increasing delays in flight.
 Failing to accurately estimate and find the pattern in
the air traffic movement
 Failing to meet the demand capacity.
 The restriction in flow control measure create air
holding problem (Cruciol et al. 2015).
 Inadequate air travel service during the high
demand hours.

Traditional air traffic
forecasting approaches
 Many regulatory Agencies use the traditional
modelling techniques developed in 1985 by
International Civil Aviation organization for traffic
forecasting (Srisaeng et al., 2015). Through review
of literature it is identified that the common
forecasting methods are
 Market Research
 Gravity model
 Simulation model
 Time Series Model
 Trend projections
 Econometric relationship forecasting

Project Aim
 The main aim of the project is to develop a model
based on machine learning that is able to
 Analyze the past data
 Find the pattern and trend data mining and machine
learning
 Using the machine learning model to forecast the
future levels of the air passenger traffic.
 Distribute the demand based on the traffic in cities.

Project Execution Plan
 The achievement of the objective requires the
execution of the developed model based on three
elements
 Collection of the data from the records and reports
 Statistical analysis of data with machine learning.
 Consideration of larger data set for the precise
estimation through Three stage models.
 Designing model for the analysis and forecasting.
 Implementation of the data mining techniques with
classification and linear regression algorithms.

Approach
 Division of the traffic data into classifiable object
 Classification of the variables with decision tree
and support vector machine algorithm.
 Combine the values of the resultant variable with
linear regression for the analysis and forecast the
future traffic with required annotations.
 Using Two models for cross verification of the
predicted result for increased accuracy.

Data collection/
analysis plan
 The data can be collected from the following areas
 Recorded dataset
 Reports
 Survey
 The collected data can be analyzed by data mining
to find pattern for determining the trends (Mao et
al. 2015).
 Analysis of the data begin by classification of the
data with pre-processing.
 Extraction of the featured data
 Applying regression for the final analysis

Historical Data
 The data recorded from 1985 to 2017 for the
domestic air traffic in Australia, reveals the
increase in the air traffic in past trends over the
past 32 years.
 With the oncoming years the demand of the airline
transport is likely to increase.
 The domestic flight service has increased far more
than the international flight service.
 In the last 10 years the average annual growth in the
passenger has been 1.9%.

Analysis
 The analysis of the data is done through following two
methods used of the forecasting the future trends and
demand
 Grouping the data with the k nearest algorithm
 Classification of the data with decision trees to sperate
the data based on labels and requirement.
 Support vector machine algorithm for finding the
estimated range for the traffic
 Linear regression model with logarithm transformation to
 Exponential smoothing of the forecast data to find the
average value of the demands
 A spatio-temporal prediction model for the forecasting
and trend prediction (Awad et al., 2017).

Proposed artefact
 The artefact proposed implement the analysis
methods discussed with the machine learning and
data mining capabilities.
 The artefact is able to apply the classification and
predicting algorithms.
 The artefact can find the pattern and trends in the
air traffic with the analyzed data.
 The artefact can also be used forecasting the
future air traffic.

About the artefact
 The artefact takes into consideration the factors that
influence the future demand like
 Number of the airports
 Population
 Income
 Fuel prices
 Airlines available at the airport
 Number of flights inbound and outbound
 With these factors the machine learning the spatio-
temporal pattern can be determined using the
gravity model along with exponential smoothing
(Awad et al., 2017).

Design of the artefact models
 The design of the model is based on the finding the
best evaluation. The extended larger range of the
variables through the data set are anticipated to the
better comprehend the factors with decision tree
(Phyoe, et al. 2015).
 Variable relation of the airline capacity, constraints,
capabilities or the airport(Gardi et al., 2015).
 The forecasting of the approaches that are used by the
models are
 Log-log model: This approach uses the linear regression
with the logarithmic conversion of the independent and
dependent variable (Chudy-Laskowska & Pisula, 2017). .
 3 stage model: The k-nearest with support vector machine
and linear regression.

Implementation
Approach
 The following approach is taken in the artefact
 Grouping the data with the k nearest algorithm
 Classification of the data with decision trees to
sperate the data based on labels and requirement.
 Support vector machine algorithm for finding the
estimated range for the traffic
 Linear regression model with logarithm
transformation to
 Exponential smoothing of the forecast data to find
the average value of the demands
 A spatio-temporal prediction model for the
forecasting and trend prediction (Awad et al., 2017).

Development models
 The development models are based on the two approaches
 2 stage log-log model
 linear regression z is
 3 stage models
 K-nearest algorithm
 Support Vector Machine
 Linear Regression

Development models
 3 stage models
 K-nearest algorithm
 Support Vector Machine
 Linear Regression
Model of Support Vector MachineK-Nearest Model

Forecasting with
intelligence
 Using advance software for the analysis of the
data based on the learned knowledge to produce
accuracy of the results. The optimal combination
of the models are used to produce the accurate
results along with identification of the disruption
of the system due to holidays, weather or events.

Result
 The result is calculated based on the cities. The
variables that are considered as the depended
variable that determine the forecast. The system
analyses the variable to find the pattern and
forecast all the available observations through the
data available.

Trends1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
20170
20,000,000
40,000,000
60,000,000
80,000,000
100,000,000
120,000,000
140,000,000
160,000,000
180,000,000
TOTAL PASSENGER
TOTAL PASSENGER Linear (TOTAL PASSENGER)
After the deregulation of domestic aviation in
1990 the private low-cost airlines have emerged
into the Australian airline industry. This has the
shown growth rate of 5% per year, in the air
travel, over the past 30 years (Busquets, Evans &
Alonso, 2015).

Responsibilities
 The responsibilities of the development of the
artefact is stated as follows
 Designer
 Developer
 Required/data collection
 Tester

Roles
 The responsibilities required be distributed in the
team and therefore the roles are
 Student 1: designing the model and algorithm
 Student 2: collecting the data
 Student 3: developing the models
 Student 4: testing the model with test case of
previously forecast data

Limitation
 The limitation of the artefact are as follows
 The model is only able to forecast and not able to
solve any air holding problem
 The can estimate the demand but cannot do
anything to manage the airport
 It requires correct data for the accurate results
 The initial run or the test can have errors as the
machine requires to learning the trends for make
accurate predictions

Comparison with the
current system
 The comparison with the current forecasting
methodology with the proposed forecasting
methodology is that the proposed system is more
accurate that ay on the previous system as they
required manual effort for making the analysis
and estimation. In the proposed system the
machine does the analysis and calculation part
based on the forecasting algorithms that are
implement with the model.

Further research
 This is this initial effort for the adaptation of the
machine learning and data mining in forecasting
the air traffic. The enhancement can evolve the
current system and the ease the management
process of the airlines to keep up with the
demand and maintain their economic growth
through servicing the high demands.

Conclusion
 The accuracy of the machine learning can
immensely benefit is the accurate and precise
forecast of the air passenger traffic. Through the
forecast the management can be prepare with the
necessary strategies to meet the demand and
maintain their economic growth.

Reference
 Busquets, J. G., Evans, A., & Alonso,
E. (2015). Application of Data
Mining in Air Traffic Forecasting. In
15th AIAA Aviation Technology,
Integration, and Operations
Conference (p. 2732).
 Phyoe, S. M., Guo, R., & Zhong, Z.
W. (2015). An air traffic forecasting
study and simulation. MATTER:
International Journal of Science and
Technology, 2(3).
 Mao, L., Wu, X., Huang, Z., & Tatem,
A. J. (2015). Modeling monthly flows
of global air travel passengers: An
open-access data resource. Journal
of Transport Geography, 48, 52-60.
 Srisaeng, P., Baxter, G., Richardson,
S., & Wild, G. (2015). A forecasting
tool for predicting Australia's
domestic airline passenger demand
using a genetic algorithm. Journal
of Aerospace Technology and
Management, 7(4), 476-489.
 Cruciol, L. L., Weigang, L., de Barros, A.
G., & Koendjbiharie, M. W. (2015). Air
holding problem solving with
reinforcement learning to reduce airspace
congestion. Journal of Advanced
Transportation, 49(5), 616-633.
 Chudy-Laskowska, K., & Pisula, T. (2017).
SEASONAL FORECASTING FOR AIR
PASSENGER TRAFIC.
 Gardi, A., Sabatini, R., Kistan, T., Lim, Y., &
Ramasamy, S. (2015, April). 4
Dimensional trajectory functionalities for
air traffic management systems. In
Integrated Communication, Navigation,
and Surveillance Conference (ICNS), 2015
(pp. N3-1). IEEE.
 Awad, Y. A., Koutrakis, P., Coull, B. A., &
Schwartz, J. (2017). A spatio-temporal
prediction model based on support vector
machine regression: Ambient Black
Carbon in three New England States.
Environmental research, 159, 427-434.

THANK YOU