=Paper=
{{Paper
|id=Vol-1185/paper6
|storemode=property
|title=Towards a Prediction Engine for Flight Delays based on Weather Delay Analysis
|pdfUrl=https://ceur-ws.org/Vol-1185/paper6.pdf
|volume=Vol-1185
|dblpUrl=https://dblp.org/rec/conf/modellierung/CabanillasCKMPP14
}}
==Towards a Prediction Engine for Flight Delays based on Weather Delay Analysis==
https://ceur-ws.org/Vol-1185/paper6.pdf
Towards a Prediction Engine for Flight Delays based on
Weather Delay Analysis
Cristina Cabanillas, Enver Campara, Claudio Di Ciccio, Bartholomäus Koziel,
Jan Mendling, Johannes Paulitschke, and Johannes Prescher∗
Institute for Information Business
Vienna University of Economics and Business, Austria
johannes.prescher@wu.ac.at
Extended abstract
Arrival performance for the United States shows that over 83 percent of flights are
actually on time. However, 17 percent delayed flights are still an indisputable high
number, having almost eight million commercial travel flights per year, only in the
US. Knowledge of the conditions leading to flight delays may be used in a monitoring
and prediction tool to diminish its impact on commercial flight operations. From a
broader perspective, we also consider multi-modal logistics chains, which involve
different modes of transportation. These modes are adopted in consecutive legs,
which have to be thus synchronized. Determining whether a delay is going to be
verified for the aircraft can thus be of advantage, in order to rearrange the overall
transportation process involving such leg. This work reports on investigations
carried out in the context of the GET (Green European Transport) Service1 project.
GET Service is an ongoing European research project, whose objective is to improve
the ecological impact and efficiency of logistics processes.
Among the possible causes of flight delays, we focus on weather. Weather is observed
throughout the world and the need to make future predictions is noteworthy.By
now, research has analysed the influence of weather on airports, on flight delays
in general, and on how a flight may be influenced by certain weather conditions.
Furthermore, models for flight delays with respect to the weather and traffic index
have been devised. However, there is little insight on the quantification of the
prediction of flight delays.
In our work, we investigate in how far weather conditions have an actual impact
on the punctuality of a flight. Following up on the insights gained in this step, we
determine categories of impacts to allow for more generalisation. Subsequently, we
∗ Corresponding author
1 http://www.getservice-project.eu/
A. Baumgraß, N. Herzberg, G. Kappel, J. Mendling, A. Meyer, S. Rinderle-Ma (eds.): Proceedings on
Inter-Organizational Process Modeling and Event Processing in Business Process Management,
Vienna, Austria, 12-06-2014, published at http://ceur-ws.org
use the categories and apply them in a prediction model. We fill the model with
historical data. Accordingly, the model and corresponding data are the foundation
for live predictions on actual flights. Our work builds upon the combination of
two data sources being weather information and flight data. The relevant weather
information is retrieved accessing the Meteorological Aerodome Report (METAR),
which is an internationally established reporting instrument for weather information.
METAR data is gathered at every airport and airfield and is usually generated
every 30 minutes. A dataset of METAR contains station meta data (which we use
to map the information to flights) as well as the information related to the weather
itself. The flight data we use is of two different types, historical data and current
flight data. We use historical data sets to analyse the cause for a delay and to
validate our prediction model. The timeframe for our dataset ranges from 2005 to
2008. We analyse flights choosing a single route, which contains both (i) enough
observable weather stations and (ii) a high amount of flights to be analysed. In
order to observe and analyse the weather at a specific point in time and the position
of each flight, we merge the collected information from METAR with the flight
information by time, date and location for flights and weather stations. We evaluate
whether there is a significant dependency between the delays of flights and certain
weather conditions occurring in the meantime. We also examine at which stage of a
flight specific weather conditions show the strongest impact. In order to conduct the
analysis of our dataset containing 869 recorded flights we use SPSS.2 In addition to
the integration of information we suggest a conceptual description of a monitoring
tool for current flights. The tool is able to predict flight delays considering the
categorised impacts in conjunction with previous flight delays and may present the
predicted delay. Basing on our data set we analyse specific weather conditions which
potentially impact flights through multiple linear regression [DS98]. The conditions
are light rain, rain, heavy rain, haze/fog, thunderstorms, light snow and snow. Our
findings indicate that there is no significant influence on delays for light rain, rain
and heavy rain. However, haze/fog, thunderstorm, light snow and snow seem to
have a noticeable impact on flight delays. We investigate these conditions to figure
out in how far they explain the delay of a flight. As a result of this analysis we
derive a factor which allows for a calculation of delay time. However, while light
snow will lead to delays when it appears close to the airport, it does not influence
the flight at all while the airplane is flying at 30,000 feet. We therefore consider
weather conditions within different stages of a flight (close to an airport or en route).
Our investigation indicates that the conditions’ impact increases significantly once
they appear closer to the airports. It identifies four weather conditions which
have a significant impact on flights. These conditions lead to different lenghts of
delay, which are considered within the linear equitation to predict the delays for
prospective flights.
A major limitation of our work is that the influence of wind has not been considered
within the analysis. Wind can be an important factor during the landing procedure
in which airplanes may not be able to reduce the speed to an optimal level due to
2 http:/www.ibm.com/software/analytics/spss/
50
tailwind. Additionally, wind at the cruise altitude may also be an important factor
for delays. The results of our analysis are strongly dependent on the quality of the
data. The historical data set of the flights includes the minutes of delays which
are based on bad weather conditions. Few cases in the data set are flights which
are stated to be delayed due to weather but do not show bad weather conditions
according to the weather stations on their way. Furthermore, in order to predict
delays, the described model obtains weather data from another prediction model,
namely the weather forecast. This forecast is afflicted with uncertainty and may
lead to deviations in our prediction. Although the predictions seem to be precise
for the flight in our scenario, the model needs to be tested with different departure
and destination airports to enhance the universal usage of the model.
Keywords: Flight Delay Prediction, Aircraft, Prediction Model, Weather, Data
Acquisition
Acknowledgements
The research leading to these results has received funding from the European Union’s
Seventh Framework Programme (FP7/2007-2013) under grant agreement 318275
(GET Service).
References
[DS98] Norman R. Draper and Harry Smith. Applied Regression Analysis (Wiley Series
in Probability and Statistics). Wiley-Interscience, third edition, 1998.
51