ABSTRACT they are one of the most rapidly expanding



Neural Networks (NN)
have acquired great success, as they are one of the most rapidly expanding
areas. Findings of this study indicated that the research areas of ANN based
applications are receiving most research attention and self-organizing map
based applications are second in position to be used in segmentation. The
commonly used models for market segmentation are data mining, intelligent
system etc. Marketing and specifically Market Segmentation (MS) is one topic
that NN can be a useful tool. This work presents an application of NN in MS. A
Case Study (CS) in the market of mobile phones is used. In this scenario, a
Mobile Company wants to predict what type of mobile phone people desire, in
order to construct and promote. For this reason, a NN was developed using Mat
lab in order to predict the most suitable mobile phone for different mobile
users. A great number of data were collected to train the NN. The decision
capabilities of this NN were evaluated. Results show the potential of NN
application in MS.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!

order now














 Artificial Intelligence (AI) is the science
and engineering of making intelligent machines, especially computer programs.
One of the most important parts of AI is Neural Networks (NN). NN is the
element that makes computers ‘think’. For this reason, NN have acquired great
success, becoming one of the most rapidly expanding areas. NN can be utilized
into many application areas such as pattern and image recognition where much
research has been done. In addition, NN can be used in business applications.
However, there are some fields that these networks have not been applied enough
yet. One topic that NN can be a useful tool is in Marketing and more
specifically in Market Segmentation (MS). In this work, the application of NN
in MS is examined. For this reason, a Case Study (CS) in the market of mobile
phones is used. In this scenario, it is assumed that a Mobile Company wants to
predict what kind of mobile phone people desire, in order to construct and
promote. A NN was developed using Mat lab in order to predict the most suitable
mobile phone for different mobile users. Several training methods were
explored. A great number of collected data were used to train the NN. After
training, the ability of the NN to decide was evaluated. The evaluation
presents satisfactory results, which demonstrate the potential of the NN usage
in Marketing and especially in MS.











 As the nature of research in ANN techniques in
segmentation are difficult to confine to specific areas, the relevant materials
are scattered across various journals. Expert systems (ES) with application and
Information system are the most common discipline for ANN research in market
segmentation. Consequently, the following online journal databases were searched
to provide a comprehensive bibliography of the academic literature on ANN
research in market segmentation:

 · Google Scholar

 · Citeceerx

· Social Science


 · Emerald Full text

 · Ingenta

· Science Direct


The literature search
was based on the query strings, ”market segmentation”, “target marketing” and
”artificial neural network”, which originally produced approximately 900
articles. The full text of each article was reviewed to eliminate those that
were not actually related to application of ANN techniques in segmentation.


The first idea for
development of this type of network was done in 1943 by the neurophysiologist
Warren McCulloch and the mathematician Walter Pitts by writing a paper about
how neurons work. The construction of one network that it can think in the same
way of a human brain, attracted many researchers because it can provide huge
capabilities. Based in simple mathematic calculations, a NN tries to simulate
learning capabilities of neurons that exist in the human brain. It depicts a
representation of a Brain Neuron. With the development of this type of network,
the idea for ‘thinking machines’ became a reality.

The input signal enters
the network, weighted biased summations are performed inside the hidden layer
and the results are exported in the output layer. The training of the NN is a
procedure where the network weights and biases are adapted every time that a
new training data is fed into the network. The input signal matrix is composed
of the signals P1, P2, PK that are received by the hidden layers. Each element
Pi is multiplied by is corresponding weight factor wi to produce weighted sums
in each node of the hidden layer. After that, every sum is biased by the value
bi to produce an output signal. Thus the output ?? can be expressed as: f (ai)
= ? (wi,i x Pi) + bi . , where bi f is the transfer or activation function. In
a generic neural network topology is presented. Inside the Neural Network There
are many areas that NN can be used such as: Pattern Recognition, Medicine,
Sports applications, Science, Manufacturing, Stocks Commodities and Futures,
Business Management, Finance and Marketing. The subject of this work is the
application of NN in Marketing and more especially in Market Segmentation.


Market Segmentation is
a part of marketing theory. It is the process of partitioning market into
smaller subgroups of potential customers with similar needs and characteristics.
The purpose of segmenting a market is to allow your marketing and sales plan to
focus on the subset of prospects that are most likely to purchase your
offering. Right segmentation will help to the highest return of marketing and
sales expenditures. Depending on whether you are selling your offering to
individual consumers or a business, there are definite differences in what you
will consider when defining market segments. There are four basic categories of
variables in MS: Geographic Variables, Demographic Variables, Psychographic
Variables and Behavioral Variables.

MS can be made with
many techniques. Neural is the one of the most recent method for segmenting
markets. Moreover, there are some other methods that are used such as
Regression Analysis and Multiple Linear Regression Analysis. When compared to
the above methods, NN are proved to give better performance, in most of the
times. There is the need to develop an application to prove so.




In the field of customer behavior, the
value of neural networks consisted in their capability to emulate the operation
of human brain and precisely estimate behavior based on product characteristics
without assumptions about relationships among input variables. Wide range of
neural networks applications in this area included customer satisfaction



The ?ber metal laminates (FML) are very important materials in many applications (especially in the aeronautic industry) due to features such
as high stiffness, low
density and long lifetime. How- ever, its various modes of deformation and its
physical structure
in multiple layers make the detection
of ?aws in early stages a dif?cult task. In
work was proposed an
automatic decision sup- port
system to help the inspector in the identi?cation of different
of defects that may appear in
FMLs. For this, a
signal processing chain which comprises two distinct
stages (feature extraction and hypothesis testing) was designed. For feature extraction, the
acquired signals were transformed to the frequency domain and processed by statistical techniques (such as PCA and ICA) in order to 
redundancy and reduce the background noise. The hypothesis testing (classi?cation) stage was performed by both linear   and neural classi?ers. The applied feature
extraction techniques were able   to reveal the underlying structure
of the data producing high discrimination ef?ciencies.
This can be observed particularly when a linear discriminator was used. In this con?guration, the ef?ciency
was considerably improved
by using PCA and ICA transformations. In the case of ICA preprocessing, the classi?cation results were quite similar
to the ones achieved when using
the nonlinear (neural) discriminators, but with considerably lower
computational requirements. Considering the neural classi?ers, the
ICA preprocessing also contributes to produce higher discrimination ef?ciencies, achieving EP =    0.998 and reducing the misclassi?cation of defect signals to     0.16%.



Costs prediction plays a crucial role in the business and product cycles.
Despite small number of found applications, neural networks offered various
advantages in costs estimation compared to traditional methods 428. With real
data, neural networks were able to extract knowledge and approximate costs
functions, or to be modified and retrained using new data.





 For this work a CS regarding mobile phone
market is considered: “A very big communication company wants to classify and
predict the consumer’s needs in purpose to construct and calculate the number
of new mobile phones. In addition, the company wants to map out the strategy of
promotion and advertise its products.” To produce such a prediction a NN can be
used. After the research on how to segment the market for the above example, a
final selection of variables was done. Fifteen final variables were selected as
questions/opinions regarding to Geographic (Region, Country, and City),
Demographic (Gender, Financial Status, Occupation, and Age), Psychographic
(Active Life, New Technology, Travelling, and Gaming) and Behavioural (New
Products, Mobile Phone, Brand and Regularity) criteria. The 15
variables/questions are used as input data for the NN. An additional 16th variable/question
about the preferred type of mobile phone corresponds to the NN output. This
variable/question is the target of each input vector for the training mode.
Collection of data is a very important and difficult task but it is essential
for training the network. In order for training data to have a real base, there
was the need for collecting data through the help of volunteers. For this
reason, an online program was used. This software can create surveys for
collection data through a web interface. The survey had three basic phases:
formulation of questions, publication of survey and collection of results. A
questionnaire combined of 16 questions was created. An email was sent to many
people through group mail addresses of Sussex University. There were a big
number of responses. More than 300 people entered the survey page and 200
people completed the questionnaire. Finally, there was a selection of answers
for training and testing the network. From the selected data, the 80% were used
for training and the rest 20% for evaluating the network.


Since the training data
were collected, they were converted into numerical by assigning each String a
number. This is necessary in order that data can be used in calculations. In
this way, each answered questionnaire was converted into a numerical matrix,
which is used as an input vector. For the presented scenario, a NN was designed
and implemented in Mat lab. Regarding the network type, the Feed-Forward Back
propagation (FFB) was chosen. FFB is the most common and effective Network type
for predicting and classifying inputs. Several training types were selected
such as the Levenberg-Marquardt optimization (TrainLM) and the Scaled Conjugate
Gradient optimization (TrainSCG). For the Adaption Learning Function the
Gradient Descent with Momentum weight and bias learning function (LearnGDM).
The implemented NN has one hidden layer. The number of neurons that were used
in the hidden layer can be varied. Different numbers of neurons were examined
in order to have the best performance. In the output layer 5 neurons were used
(as many as the outputs-types of mobile phones). Since the system is nonlinear,
the transfer function that had the better performance was Tansig (tangential)
for the hidden layer and Logsig (logarithm) for the output layer. After the NN
creation, the training of the NN is a very important issue. The training data
vectors were used with both training methods and with different number of
neurons in the hidden layer. Three matrixes are needed for training the
network: the input matrix, the target matrix and the input range matrix. For
the input matrix, the results of the 14 questions were used (finally, the city
question was not included as not applicable). Concerning the target, the 16th
question (type of preferred mobile phone) was utilized. Finally, a matrix
containing the range of each element of the input matrix was constructed.


For the evaluation of
the network, 40 data vectors were used. The answers (regarding the type of the
mobile phone) produced by the network were compared with the real ones. The
results of the evaluation are presented in Table 1 and 2 for TrainSCG and
TrainLM method respectively for different number of neurons. The first column
contains the number of neurons, the second the training performance (Mean
Squared Error), the third the number of epochs (training iteration) and the
forth the succession percentage. As it can be seen, for the TrainSCG approach,
the success is ranged from 26.7% to 50%. The 50% correct outputs were produced
by a network using 35 neurons. In addition, the test shows that the number of
neurons in the network is a very important parameter but it is not proportional
to the success percentage. The second method that is presented (TrainLM)
produced better results. As it can be seen, correct answers were more than
33.3% in all different number of neurons. The higher number of correct answers
was reached for 37 neurons in the hidden layer, which is 70% or 28 over 40
giving great efficiency for this kind of network. However, this method involves
higher processing complexity.

There are some comments
about the results of the Evaluation: Results can be assumed successful since a
big percentage of correct answers were accomplished. Even for the TrainSCG
method, the results were satisfying enough. TrainLM proved to be better in
comparison to the TrainSCG method. The number of neurons that gave best results
was close for both training methods.

(35 neurons 50% for TrainSCG,
37 neurons 70% for TrainLM) The number of neurons is of very importance for the
training of the network. It is not possible to know the correct answer of the
neurons before the test. The training of the network was proved a
time-consuming procedure. Especially for TrainLM method.

















 In this work, the application of NN in MS is
proposed. In order to examine this, a case study regarding the mobile phone
market is presented. Several NN designs and training methods were investigated.
Two training methods produced satisfactory results dictating the potential of
NN usage in MS, which can be a powerful tool in the Marketing Science. The
presented work is an example to show the methodology for future research and
implementation. This method can be used in a very big variety of applications of
Marketing. It can be used in every market that has to be segmented. In order to
improve further the performance of such application several issues have to be
taken into account: Selection of questions (which are used as training inputs)
is critical. Different formulations for the questions may give different
results. Range of possible answers also affects the results. Variation of the
statistical sample is very significant. The amount of training data is
important. A greater database can give better results. The network proved to be
sensitive. The size and type of NN as well as the training method is critical.
The application of NN in MS requires a good background in many different
scientific fields such as marketing, neural network, sampling and statistics.
The capability of neural networks to take decisions is very hopeful for
developing more applications on the field of market segmentation. It provides
to the future scientist a great area for research and development.

ANN can be applied to
many marketing problems which could be tackled previously by statistical
analysis only. Typical problems turn out to be market segmentation tasks and
more dominantly market response modelling, classification of consumer spending
patterns; new product analysis; identification of customer characteristics;
sale forecasts, targeted marketing; and modeling the relationship between
market orientation and performance. Most of cited papers explicitly compare
ANN- approaches to traditional methods including discriminate analysis for
classification tasks and estimations of market response functions by multiple
regression analysis.







1.      Anil,
C., Carroll, D., Green, P. E., & Rotondo, J. A. (1997). A feature-based
approach to market segmentation via overlapping K-centroids clustering. Journal
of Marketing Research, Vol. XXXIV (August), 370–377

2.      Active8
(2005), Predictive Technologies, The History and Application of Artificial
Neural Network, Human Resources White Paper. Bloom Jonathan Z (2004), Market
Segmentation, A Neural Network Application, doi:10.1016/j.annals.2004.05.001.
Brassington Frances and Pettitt Stephen (2003), Principles of Marketing, 3rd,
Printed in Italy, ISBN 0-273- 65791-7.

3.      Baesens
B., Viaene S., Poel D. V. den, Vanthienen J., Dedene G., (2002). Bayesian
neural network learning for repeat purchase modelling in direct marketing,
European Journal of Operational Research, 138( 1, 1), pp. 191-211

4.      DeTiennen
Kristen Bell, Lewis Lee W (2003), Artificial Neural Networks for the Management
Researcher: The State of the Art, Marriott School of Management Brigham Young
University. Euro regional Center for Democracy (CED) (2005), Specific Issues
Relevant For SME Cross-Border Business Development, Timisoara Romania Haykin,
Simon (1999), Neural Networks, A Comprehensive Foundation, 2nd edn, Printed in
USA, ISBN 0-13- 273350-1.

5.     Chan
C. C. H., (2008). Intelligent value-based customer segmentation method for
campaign management: A case study of automobile retailer, Expert Systems with
Applications, 34(4), pp. 2754-2762

6.      Balakrishnan,
P.V., Cooper, M.C., Jacob, V.S., Lewis, P.A.,1994. A study of the classi®cation
capabilities of neural networks using unsupervised learning: A comparison with
k-means clustering. Psychometrika 59, 509±525.

7.      Balakrishnan,
P.V., Cooper, M.C., Jacob, V.S., Lewis, P.A.,1996. Comparative performance of
the FSCL neural net and K-means algorithm for market segmentation. European
Journal of Operational Research 93, 346±357.

8.      Bishop,
C.M., 1995. Neural Networks for Pattern Recognition. Oxford University Press,

9.      Bridle,
J.S., 1990. Training stochastic model recognition algorithms as networks can
lead to maximum mutual information estimation parameters. In: Touretzky, D.S.
(Ed.), Advances in Neural Information Processing Systems 2. Morgan Kaufmann,
San Mateo, CA, pp. 211±217.

10.     Cheng,
B., Titterington, D.M., 1994. Neural networks: A review from a statistical
perspective. Statistical Science 9, 2±54.Davies, D.L., Bouldin, D.W., 1979. A
cluster separation measure. IEEE Transactions on Pattern Analysis and Machine
Intelligence 1, 224±22

11.  w11x
J. Jarrett, Business Forecasting Methods, Basil Blackwell, 1991.

12.  w12x
A. Kaufmann, M.M. Gupta, Introduction to Fuzzy Arithmetic, North-Holland,
Amsterdam, 1985.

13.  w13x
A. Kumar, V.R. Rao, H. Soni, An empirical comparison of neural network and
logistic regression models, Marketing Letters 6 _4.  1995. 251–263.

14.  w15x
R.J. Kuo, P.H. Cohen, Manufacturing process control through integration of
neural networks and fuzzy model, Fuzzy Sets and Systems_1998. _to appear..

15.  w17x
C.C. Lee, Fuzzy logic in control systems: fuzzy logic controller-parts I and
II, IEEE Transactions on Systems, Man, and Cybernetics 20_2. _1990. 404–435.

16.  w18x
G.S. Le Vee, The key to understanding the forecasting process, Journal of
Business Forecasting, Vol. 11, Issue 4, Winter _1992–1993.12–16

17.  w25x
Z. Tang, C. Almeida, P.A. Fishwick, Times series forecasting using neural
networks vs. Box–Jenkins methodology, Simulations,

18.  w23x
G.G. Meyer, Marketing research and sales forecasting at Schlegel corporation,
Journal of Business Forecasting 12 _2. _1993. 22–23.


19.  w22x
R.P. Lippmann, An introduction to computing with neural nets, IEEE ASSP
Magazine, April _1987. 4–22.

20.  w21x
C.T. Lin, A neural fuzzy control system with structure and parameter learning,
Fuzzy Sets and Systems 70 _1995. 183–212.


22.  w20x
C.T. Lin, Y.C. Lu, A neural fuzzy system with linguistic teaching signals, IEEE
Transactions on Fuzzy Systems 3 _2. _1995. 169–189.


24.  Thesis,
Department of Industrial Engineering, Pennsylvania State University, 1994.

25.  w14x
R.J. Kuo, Multi-sensor integration for intelligent control of machining through
artificial neural networks and fuzzy modeling, Ph.D.

26.  w19x
C.T. Lin, C.S.G. Lee, Neural-network-based fuzzy logic control and decision
system, IEEE Transactions Computer C 40 _12. _1991.w16x G. Lachtermacher, J.D.
Fuller, Backpropagation in time-series forecasting, Journal of Forecasting 14
_1995. 381–393.