IJACM , International Journal of Arts Commerce & Management, VOL.01 No.

01, November 2011

On applying NeuroNeuro- Computing in EE-com Domain

Asif Perwej
[email protected]
Singhania University, Rajsthan, India
Abstract
Prior studies have generally suggested that
Artificial Neural Networks (ANNs) are superior to
conventional statistical models in predicting
consumer buying behavior. There are, however,
contradicting findings which raise question over
usefulness of ANNs. This paper discusses
development of three neural networks for modeling
consumer e-commerce behavior and compares the
findings to equivalent logistic regression models.
The results showed that ANNs predict e-commerce
adoption slightly more accurately than logistic
models but this is hardly justifiable given the added
complexity. Further, ANNs seem to be highly
adaptive, particularly when a small sample is
coupled with a large number of nodes in hidden
layers which, in turn, limits the neural networks
generalisability.
Key words: Neural networks, E-commerce, consumer
behavior

1.

Introduction

During recent years, researchers have utilized

Artificial Neural Networks (ANNs) to solve a wide
array of business and marketing research problems
including regression [7, 12], classification [29, 31]
and clustering/grouping, with some promising
results. The ANNs seem to be particularly
amenable to classification problems where the
outcome variable encompasses two or more
categories [30]. Earlier studies which compared the
predictive performance of discriminate analysis
with that of neural networks all concluded that
ANNs are superior analytical tools when dealing
with classification problem [28, 29, 31]. More
recently, Shih and Fang (2005) [27] observed that
ANNs were more accurate than both Discriminate
Analysis and Decision Tree Analysis in predicting
online shopping intention.
Despite all the favorable findings, the ANNs
literature remains mixed as there is also some
discouraging evidence (see for e.g., [15]). Given
the inconsistency in the literature, this paper
intends to provide further empirical evidence by
comparing logistic regression and neural networks
in the context of consumer e-commerce. The
remainder of the paper is organized into the
following four sections. First we review the
consumer e-commerce literature followed by a
section on developing three ANNs. The third
section discusses the performance of neural
networks and compares it to logistic regression.

Finally, the concluding section highlights some key

findings emerged from our analysis.

2.

Related Research Review

In seeking to explain the adoption of e-commerce

by consumers, researchers have employed various
theories and conceptual frameworks such as the
Technology Acceptance Model [23], domainspecific innovativeness [11], shopping orientation
[4], [9] and consumers perceptions of online risk
[10,19]. Demographic variables have also been
widely used to predict consumer adoption behavior.
Past results indicate that adopters of e-commerce
tend to be young [8, 13]. Also, previous research
examining the effect of sex revealed that males are
more likely to conduct online transactions than
females [4, 8, 16, 17, 20, 14]. It is also found that
household income positively affects adoption of
online shopping [17], had a positive impact on
frequency and amount spent online [6, 16, 20] as
well as predicting intention to shop via the internet
[14]. In addition, empirical findings indicate that
better educated individuals are more likely to
favour online shopping [14, 17], purchase more
frequently and spend more online [6].
The usefulness of demographics as predictor of
consumer buying behavior has also been tested in
context other than e-commerce with promising
results [1, 5, 18 22, 24, 25]. Therefore, based on the
past empirical findings, the current study uses nine
demographic variables, cited by [26] as the most
commonly used demographics, to predict adoption
of consumer e-commerce.

3.

Implementing neural networks

Essentially, artificial neural network is computer

software which attempts to mimic biological neural
systems. Like its biological counterpart, ANN has
ability to learn complex patterns from data and
subsequently generalize them into new contexts. It
is suggested that ANNs are particularly useful
when data exhibit non-linear and interaction
relationships [15]. Details of ANNs are beyond the
scope of this paper, however, for an excellent nontechnical discussion, consult [21, 30].
Three feed forward Multi-layer Perceptron
Networks [3] were developed to discriminate the
adopters of consumer e-commerce from the nonadopters. The Networks share the same input and
output layer but differ in terms of the number of
processing elements (1-node, 5-node and 10-node
network) in the hidden layer. Figure 1 illustrates

IJACM , International Journal of Arts Commerce & Management, VOL.01 No. 01, November 2011

the network with 10 nodes in the hidden layer.

There are nine nodes in the input layer which
correspond to the nine independent demographic
variables. Previous studies suggested that the
number of processing elements in the hidden layer
is positively associated with the predictive
performance of the networks [3]. However, there is
a downside to this. A greater number of nodes in
the middle layer increases the so-called network
capacity which, in turn, limits the networks
generalisability. It was mainly due to this reason, as
well as gaining an in-depth understanding of
ANNs behavior under different circumstances,
that we fitted three different types of networks to
the data (i.e., with one, five and ten nodes in the
hidden layer). The output layer consists of two
nodes, one being for adopters and the other for
non-adopters which essentially are the same as
outcome variables in the logistic regression
analysis. Finally, the epoch or number of iterations
of the networks was set to 200 which is the
maximum iteration available in our software (i.e.,
Nuclass 7.06)

Figure-1 Neural architecture for adoption of consumer behavior

4.

Results and discussions

Data for this study are drawn from General Social

Survey 2002 (Australian Bureau of Statistics,
2003). The dataset contains information on 15,510
individuals actual e-shopping, e-government, ebanking and e-share trading adoption which were
treated as dependent variables and coded as 1=
adopter and 2 = non-adopter. To avoid unequal cell
size problem, we have created 20 sub-samples with
an equal number of adopters and non-adopters (this
later one selected randomly from corresponding
non-adopter respondents) for each sample.
Table 1 depicts the results of Artificial Networks.
Two major findings have emerged form this
analysis. First, the networks exhibit good predictive
power regardless of the number of processing
elements in the hidden layer. The hit ratio ranges
from 67.1 percent to 92 percent which should be
compared to the Maximum Chance Criterion
(MMC) of about 50 percent. Second, the networks
with higher number of nodes in the middle layer
seem to perform better then the networks with
lesser nodes in the hidden layer. For example, on
average, the networks with 10 nodes have correctly
classified 78 percent of the respondents in terms of
adoption of consumer e-commerce whereas
comparable figure for the networks with five and
one node in the hidden layer were 76 and 72
percent, respectively. Comparing Results of
Artificial Neural Networks and Logistic Models
ANOVA analyses were performed to compare the
results of the Neural Networks with performance of
logistic regression analysis which were reported by
[24]. Note that, to facilitate comparability of the
results, the same 20 samples which were used by
[24] for logistic regression analysis were employed
to conduct the ANN analysis.

IJACM , International Journal of Arts Commerce & Management, VOL.01 No. 01, November 2011

Table-1

Ref.
1
2
3
4
5
6
7
8
9
10
11
12
13
14

15
16

17

18

19

20

Dependent variables
e-banking
e-share trading
Online Shopping
Food and grocereries
Alcohol
Toys
Videos or Dvd's
Music or CD's
Books or Magazines
Financial services
computer hardware
or peripherals
Computer software
Clothing or shoes ,etc
Sporting equipment
Travel or
Accommodation
Ticket to
entertainment or the
cinema
Other goods or
services
E-government
Electronic lodgment
of tax returns
Electronic lodgment
of applications or
claims
for benefits
Electronic lodgement
of application for
permit:etc
Electronic lodgement
of bill payments
(e.g rate:and car
registration )
Mean

Hit Ratio
With 5 node
With 10 node
in
in
hidden layer
hidden layer
71.2
71.7
73.9
76.5

Sample
size
7124
988

With 1 node
in
hidden layer
69.9
73

575
188
253
547
786
1445
678

72.9
78.2
72.7
73.9
68.3
70.4
73.2

76.2
86.7
81.8
77.7
75.7
72.8
76.1

446
997
675
277

76.9
69.9
71.1
76.2

2412

Maximum
chance
criterion
50.3
50.2

Logistic
regression
Hit Ratios
70.5
74.5

80
92
80.2
79.5
76.7
74
79.4

50.3
50.5
50.2
50.3
50.5
50.2
50.7

73.9
85.1
75.9
73.3
71.8
69.9
73.3

80.5
72.6
74.8
79.8

83.6
73.8
77
81.6

50.4
50.1
50.1
50.5

75.3
69.9
73.6
75.3

72.4

73.3

73.8

50

73

1217

73.2

77.2

76.2

50.5

74.5

937

67.1

68.4

71.2

50.6

67.7

1320

67.5

71.8

71.1

50.5

68.5

310

72.6

78.7

81.6

50.3

74.8

554

70.4

74.9

78

50.3

70.6

3499

70.1
72

71.4
76

73.3
78

50.3
50

69.9
73

Table-2
Mean
Logistic regression analysis
Artificial Neural Network Analysis
Hidden layer with 1 node
Hidden layer with 5 nodes
Hidden Layer with 10 nodes

Further, to be consistent with the results of the

logistic models, all the data were submitted to
training of the networks and, as such, no cross
validation (i.e., testing) was conducted. As can be
seen from Table 2, no statistically significant
discrepancy was detected between the result of

Standard
Deviation

f-value

73.13

3.82

71.99
75.78
77.56

2.95
4.29
5.01

1.095
4.258
9.901

p-value

0.302
0.046
0.003

logistic regression (mean hit ratio=73.13 percent)

and the result of artificial neural network with one
node in middle-layer (mean hit ratio=71.99
percent). However, the average hit ratio of the
networks with five and ten nodes in the middle
layer (75.78 percent and 77.56 percent,

IJACM , International Journal of Arts Commerce & Management, VOL.01 No. 01, November 2011

respectively) was significantly higher than mean hit

ratio of logistic regressions.

2.

Our findings corroborate the results of previous

studies suggesting that neural networks may
outperform conventional statistical models, such as
logistic regression in this case. Two key findings
emerged from this study. First, the increased
accuracy is marginal and may not outweigh higher
complexity involved with designing and operating
ANNs. In the best case scenario, the ANNs added
only about 4.5% to the accuracy of logistic models.
Second, the performance of the neural networks
seems to be affected by ANNs architecture. We
found, not surprisingly, that networks with iteration
rate of 200 have yield better hit ratios than
networks with, for example, only 50 iterations.
Also the number of processing element in the
middle layer exerted a positive influence on the
networks accuracy. The impact of sample size on
the networks performance is rather complex and
not very well documented. Our observation
indicates that when there is a limited number of
patterns in place (i.e., small sample size), the
networks tend to memorise extant patterns amongst
the data and consequently return higher predictive
accuracy. Importantly, this particular characteristic
of ANNs became more evident as the number of
nodes in the hidden layer increases. As such, it
would appear that small sample size magnifies the
problem associated with including large number of
processing elements in the hidden layer via
increasing network capacity and adaptability.

3.

No doubt, ANNs have brought about some great

advances.
Compared
with
conventional
multivariate analysis, ANNs better handle
multicollinearity, non-linear functions, noisy and
missing data. One major problem with ANNs,
among others, is the possibility of infinite network
architecture in terms of network types, iterations,
number of nodes and layers and so forth. This
makes the outcomes of ANNs rather arbitrary.
Given the indispensable role played by ANNs
architecture in determining the networks
outcomes, we suggest that future research
endeavours in this field should focus on developing
guidelines, similar to the rules of thumb used in
conventional statistical methods. This will help
researchers in the selection of the appropriate
network features in order to obtain optimal and
comparable model performance.

References
1.

Assael, H., Poltrack, D. F., 1994. Can

demographic profile of heavy users serve
as a surrogate for purchase behavior in
selecting TV programs? Journal of
Advertising Research 34 (1), 11-17.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

Australian Bureau of Statistics., 2003.

Household
Use
of
Information
Technology, ABS, Canberra.
Berry, M.J.A., Linoff, G.S., 2004. Data
Mining Techniques: For Marketing, Sales,
and Customer Relationship Management.
Wiley, Indianapolis.
Brown, M., Pope, N., Voges, K., 2003.
Buying or browsing? An exploration of
shopping orientations and online purchase
intention. European Journal of Marketing
37 (11/12), 1666-1684.
Baumann, C., Burton, S., Elliott, G., 2005.
Predicting actual consumer behaviour in
banking. Proceedings of the 36th Annual
Meeting of Decision Sciences Institute
(WDSI). San Francisco, 19-22 November
2005
Burroughs, R.E., Sabherwal, R., 2002.
Determinants
of
retail
electronic
purchasing: A multi-period investigation.
INFOR 40 (1), 35-56.
De Groot, C., Wurtz, D., 1991. Analysis
of
univariate
time
series
with
connectionist nets: A case study of two
classical examples. Neuro Computing 3,
177-192.
Donthu, N., Garcia, A., 1999. The Internet
shopper. Journal of Advertising Research
39(3), 52-58.
Fenech, T., OCass, A., 2001. Internet
users adoption of web retailing: User and
product dimension. Journal of Product and
Brand Management 10 (6), 361-381.
Gefen, D., Karahanna, E., Straub, D.W.,
2003. Trust and TAM in online shopping:
An integrated model. MIS Quarterly 27
(1), 51-90.
Goldsmith, R.E., 2001. Using the domain
specific innovativeness scale to identify
innovative internet consumers. Internet
Research:
Electronic
Networking
Application and policy 11 (2), 149-158.
Hruschka, H., 1993. Determining market
response functions by neural network
modelling:
A
comparison
with
econometric techniques. European Journal
of Operations Research 66, 27-35.
Joines, J.L., Scherer, C.W., Scheufele,
D.A., 2003. Exploring motivation for
consumer web use and their implication
for e-commerce. Journal of Consumer
Marketing 20(2), 90-108.
Kim, E.Y., Kim, Y.K., 2004. Predicting
online purchase intention for clothing
products. European Journal of Marketing
38(7), 883-897.
Koh, H.C., Low, C.K., 2004. Going
concern prediction using data mining
techniques. Managerial Auditing Journal
19 (3), 462-476.

IJACM , International Journal of Arts Commerce & Management, VOL.01 No. 01, November 2011

16.

17.

Korgaonkar, P.K., Wolin, L.D.,1999. A

Multivariate analysis of web usage.
Journal of Advertising Research 39 (2),
53-68.
Koyuncu, C., Lien, D., 2003. E-commerce
and consumers purchasing behavior.
Applied Economics 35, 721-726.

18.

Hortman, S.M., Allaway, A.W., Mason,

J.B., Rasp, J., 1990. Multisegment
analysis of supermarket patronage. Journal
of Business Research 21, 209-223.

19.

Liebermann, Y., Stashevsky, S., 2002.

Perceived risks as barriers to internet and
e-commerce usage. Qualitative Market
Research: An International Journal 5 (4),
291-300.
Lightner, N.J., Yenisey, M.M., Ozok,
A.A., Salvendy, G., 2002. Shopping
behaviour and preferences in e-commerce
of Turkish and American university
students: Implications from cross-cultural
design.
Behaviour
&
Information
Technology 21 (6), 373-385.
Metaxiotis, K., Psarras, J., 2004. The
contribution of neural networks and
genetic algorithms to business decision
support: Academic myth or practical
solution? Management Decision 42 (2),
229-242.
McCann, J.M., 1974. Market segment
response to the marketing decision
variables. Journal of Marketing Research
XI (Nov.), 399-412.
McKechnie, S., Winklhofer, H., Ennew,
C., 2006. Applying the Technology
Acceptance Model to the online retailing
of financial services. International Journal
of Retail and Distribution Management
34(4/5), 388-410.
Naseri, M.B., Elliott, G., 2007. Comparing
the predictive utility of demographics to
alternative
segmentation
schemes.
Proceedings of the Academy of Marketing
Conference. London: Kingston Business
School, Kingston University
Novak, T.P., MACEvoy, B., 1990. On
comparing
alternative
segmentation
schemes: The List of Values (LOV) and
Values and Life Style (VALS). Journal of
Consumer Research 17 (June), 105-109.
Pol,
L.G.,
1991.
Demographic
contributions
to
marketing:
An
assessment. Journal of the Academy of
Marketing Science 19 (1), 53-59.
Shih, Y., Fang, K., 2005. Customer
defecations analysis: An examination of
online bookstores. The TQM Magazine 17
(5), 425-439.
Subramanian, V., Hung, M.S., Hu, M.Y.,
1993. An experimental evaluation of

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

neural networks for classification.

Computers and Operations Research 20
(7), 769-782.
Tam, K.Y., Kiang, M.Y., 1992.
Managerial
applications
of
neural
networks: The case of bank failure
predictions. Management Science 38(7),
926-947.
Venugopal, V., Baets, W., 1994. Neural
networks and statistical techniques in
marketing research:
A
conceptual
comparison. Marketing Intelligence &
Planning 12 (7), 30-38.
Yoon, Y., Swales, G., Margavio, T.M.,
1993.A comparison of discriminant
analysis versus artificial neural networks.
Journal of Operational Research Society
44 (1), 51-60.