Flight Revenue Information Support System for Ethiopian Airlines
No Thumbnail Available
Date
2000-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Ethiopian Airlines is a profit-oriented business organization whose objective is to provide
the maximum value to its customers, consistent with the need to make some return on each
transaction. One of the major primary activity at the airline is Sales. In addition, because
conU11erci al organizations only survive by identifying and satisfying the market, Marketing
Services is also regarded as a major primary activity.
This study focuses on the revenue process within the Sales and Marketing operation. In
particular, it aims at understanding the critical business functions and processes involved
in the flight revenue process, to identify and assess the avai labi lity of revenue data
elements and develop a model for Ethiopian Airlines that wi ll support information on
revenue realized by flight and forecast revenue by flight; accurately and timely.
Ethiopian Airlines has numerous state-of-the-art application systems and as a result retains
a vast amount of data in its different databases. However, it has fai led to make good use of
this data and has not been ab le to lise it to create competitive advantage. As a result, the
revenue information model has been developed using data mining techniques. In
particular, the neural network model was used to train, test, validate and develop the
prototype model. The ultim ate objective being to find out the suitability of data mining
applications to the Ethiopian Airl ines problem.Since the scope of the study is limited to a single organization, the major method that has
been used to assess revenue information needs of users is case study; implemented
through interv iews (planned discussion), questionnaires, observation and document
analysis.
After reviewing the various areas that are affected by the Sales and Marketing operation;
Sales, Scheduling, Pricing, Revenue Management and Airport Operations have been
identified as the critical functions in the revenue process. As a result the foc us of the study
has been on these functions.
Survey results reveal that of the 5 most important infonn ation required by the concerned
airline managers, revenue related information ranks on top with 31 % of respondents
ranking it first. In addition, 84% of respondents rate fl ight revenue information as either
one of the most or the most critical infOim ation, 88% as either very or extremely strategic,
and 94% as one that would provide opportunity to gain competitive advantage.
The major revenue related data elements identified during the study are advanced booking
data, post departure data, schedule data, and revenue data. These revenue related data
elements are available within the existing system, but are scattered in the various
application systems. Over one year's historic advance booking data is available, over two
years' post-departure data is avail able, and historical fl ight revenue data since April, 1997
is available.After selecting a suitable software to build a revenue information model, the revenue
related data elements identified were collected for 8 flights and a comprehensive testing
was conducted. The test included 6 different experiments using the back propagation
network and radial basis fll11ction neural network models, 3 different sets of independent
variab les and a multitude of trai ning parameters. The experiments produced 327 different
models which were compared and evaluated and finally one was selected to represent the
revenue information model. The developed model, with an average of33-37% error rate, is
only a preliminary or initial step towards, hopefully, more detailed work in this area.
I am confident that through a selection of more fields and with more hi storical data, the
error may be able to be reduced to users' requirement of 5-10%. It is, therefore, my belief
that this research has some contribution to further research in this area. It has been ab le to
successfully demonstrate that data mining applications can be an alternative approach to
build information systems; especially for complex problems having vast amount of data
and high interaction among non linear variables. Others can pursue similar research using
different types of data mining applications, including other neural network models. I hope
that some of the problems I encountered and the methodologies I used will help to shed
light and guide others undertaking similar studies.