International Journal of Engineering and Technical Research (IJETR)

ISSN: 2321-0869, Volume-2, Issue-5, May 2014

Comparative study of Institute based ERP based on

ANFIS, ANN and MLRA
Dr. Rajeev Gupta, Sarita Chouhan, Neha Bhuria
flag impending failures in ERP implementations? This is
Abstract Now a days, there is an essential need for efficient the research question we seek to answer in this paper. This
ways of continuous assessment, identifying shortcomings and research has developed a method of predicting User
improving system performance. On one hand, the quality of Satisfaction, a key measure of ERP project success using ex
ERP systems is related to the user satisfaction. On the other ante causal factors as predictors.
hand, measuring humans satisfaction is intermingled by
This study consolidates and extends an earlier study which
uncertainty and vagueness. The main objective of this study is
to identify the antecedents of end-user satisfaction with an gathered data from a cross section of business
Enterprise Resource Planning (ERP) system, in the context of a organizations that had implemented ERP systems in the last
educational institute. That is why ordinary statistical analysis three years and developed and tested a measurement model
is not necessarily efficient in this context. This motivated us to for causal factors for success. Data was collected, using a
use fuzzy logic methods in assessing the effectiveness of ERP. structured questionnaire, on Cri
The results establish that ANFIS is able to predict outcome -tical Success Factors (CSFs), identi-
well with an error (RMSE) of 0.2945 and outperforms ANN fied in literature as being causal for the success of an ERP
and MLRA with errors of 0.85 and 0.86 respectively. This
impleme-ntation and overall User Satisfaction , a key
study is expected to provide guidelines academia to predict
indicator of the success. Respondents to our questionnaire
ERP outcomes and thereby enable corrective actions to
redirect ailing projects. represented different user cohorts: Strategic Users,
Technical Users and Operational Users. The validity and
reliability of the measurement model and its innate value as
Index Terms ANFIS, ERP Implementation Outcome, a predictor of ERP success was established using Structural
Antecedents, Prediction. Equation Modeling (SEM) with LISREL 8.7.
In the present study the data from the earlier study was used
I. INTRODUCTION to develop predictive models for ERP implementation
The track record of successful IT projects of which outcomes measured in terms of User Satisfaction.
Enterprise Resource Planning (ERP) is a subset projects Three prediction techniques, Multiple Linear Regression
remains poor. The latest CHAOS study of the Standish Analysis (MLRA), Artificial Neural Networks (ANN) and
Group reports a marked decrease in IT project success rates, Adaptive Neuro Fuzzy Inference System (ANFIS) were
with only 32% succeeding in on time and on budget tested. Of the three ANFIS was found to be significantly
delivery with required features and functions. better in predicting User Satisfaction of an ERP project.
44% were delivered late or over budget, and/or with less This paper is organized as follows: Section 2 presents the
than the required features and functions. 24% were can- literature review and establishes the need and relevance of
celled prior to completion or delivered and never used. This this research work. Section 3 outlines the method used in the
is worse than the figures of about decade back as observed research. This section also explains different prediction
by Robey in 2002: About half of ERP projects fail to techniques with specific emphasis on ANFIS. Section 4
achieve anticipated benefits. presents the results of the modeling and compares the
Information systems (IS) project failures often encounter results of the various techniques used. Section 5 concludes
project escalation defined as a continued commitment to a the paper with the direction for continuing research.
failing course of action despite uncertainty surrounding the
likelihood of goal attainment. Escalation research lists II. METHODOLOGY
issues that cause escalation and suggests strategies for The conceptual model underlying the present study is
de-escalation which includes abandoning or redirecting given in Figure 1.
the project. While these are acceptable as reactive steps a
proactive approach of predicting impending failures, would
be invaluable as one could then attempt to forestall or at least
redirect the project far better.
The essence of proactive control is having predictive
capabilities. The challenge is to move from the diagnosis of
the source of past problems to the prediction and forecasting
of potential problems in new projects. Can a robust, easy
to use and reliable predictor be developed that would red

Manuscript received May 16, 2014.

Figure 1. Conceptual model.

195 www.erpublication.org
 Comparative study of Institute based ERP based on ANFIS, ANN and MLRA

Through an earlier empirical study data was collected from a adaptive learning algorithm and use an information
cross section of around 12 students and 120 respondents processing system composed of a large number of
representing three user cohorts: Strategic, Technical & interconnected processing elements (neurons) working in
Operational responded to a pre-tested and validated (for tandem. Neural networks are made of basic units arranged in
content validity) structured questionnaire. Respondents rated layers. A unit collects information provided by other units (or
the CSFs present in their organizations during ERP by the external world) to which it is connected with weighted
implementation. The CSFs list used for this research was connections called synapses. These weights, called synaptic
drawn from prior research, and confirmed by an expert panel weight multiply (i.e.amplify or attenuate) the input
as relevant for the current context. The CSFs were also information. A positive weight is considered excitatory, a
validated as relevant as per Structuration, negative weight inhibitory. One of the most popular
Expectations-Confirmation, Lewins Change and Agency architectures in neural networks is the multi-layer perceptron
theories. Responses were captured on a Likert scale with end which is illustrated in Figure 2. Learning happens through a
values of 5 = Completely Agree and 1 = Completely methodology of continuously altering the weights to achieve
Disagree. From the same set of respondents their overall closer and closer values to desired output. One algorithm that
satisfaction, a measure of Success of the ERP project was also performs this is known as the back propogation algorithm.
captured on a seven point Likert Scale with end values of 7 = The back propagation algorithm is a generalization of the
Completely Satisfied and 1 = Completely Dissatisfied. least mean squre algorithm . The network weights are
modified to minimize the mean squred error between the
III. ALGORITHM REQUIREMENT desired and the actual output of the network.The network is
trained using a training data set where the inputs and output
Three different prediction techniques were used:- Value are Known. After the tranning is completed, the
1) Multiple Linear Regression Analysis (MLRA) weights are frozen and the models can be used be prediction
2) Artificial Neural Network (ANN) or outputs for new set of inputs Values.
3) Adaptive Neuro-Fuzzy Interference System (ANFIS)
In all cases about 70% data was used to build/train the model.
The balance 30% of data was used for testing the model.
Each of these is discussed in the following paragraphs.

A. Multiple Linear Regression Analysis (MLRA)

Linear least squares regression analysis is still the most

common technique used, as observed in the literature. Being
a pure statistical technique MLRA has a few important
underlying assumptions. These are 1) linearity-the
assumption that the predictor variable is linearly related to
the dependent variable, 2) no multicolloinearitythe
individual predictors are not correlated to each other, 3) no
heteroscadacity-the error variances of the predictor
variable are constant across the range of data. These
conditions make the use of MLRA restrictive especially when
modeling issues related to human judgment where C. Adaptive Neuro-Fuzzy Inference Systems (ANFIS)
multicollinearity and heteroscadacity are sometimes
unavoidable. However, despite its limitations MLRA is an While ANN is a good technique that emulates the way a
established technique and this study compares the results of human brain makes a judgement, a limitation is the way t
MLRA with results obtained from other prediction handles the input data. In the case of human reasoning input
techniques. data need not always be crisp but could have linguistic labels
like small, high etc. Also, the response to data need not
B. Artificial Neural Networks (ANN) always follow a strict Yes-No rule but could have a range of
responses across a continuum.
Such a pattern of responses is referred to as the membership
The most common model-building technique identified in
function and such reasoning is called fuzzy reasoning. A
the literature as an alternative to MLRA is back- propagation
fuzzy inference system using fuzzy rules can model
trained feed-forward neural networks often referred to simply
qualitative aspects of human behavior. This was first
as back-propagation networks. ANNs are complex and
explored by Takagi and Sugeno and has since been used in
flexible nonlinear systems with the ability to deal with noisy
numerous applications involving pre-dictions. Fuzzy
or incomplete input patterns, high fault tolerance, and the
inference systems are composed of five functional blocks as
ability to generalize from the input data. Neural networks
given in Figure 3. These are 1) a rule base containing a
excel at applications where pattern
number of if-then rules 2) a database which defines the
recognigation is important and precise computational
membership function, 3) a decision making interface that
answers are not required. ANN works on the principle of an
operates the given rules 4) a fuzzification interface that

196 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-2, Issue-5, May 2014
converts the crisp inputs into degree of match with the
linguistic values like high or low etc., and 5) a de Layer 3:The nodes in this layer are also fixed nodes. It
fuzzification interface that reconverts to a crisp output. calculates the ratio of a rule firing strength to sum of the
firing strengths of all the rules.

Layer 4:Each node in this layer is an adaptive node, whose

output is simply the product of the normalized firing strength
and a first-order polynomial(for a first
Order Sugeno model ).
Thus, the outputs of this layer are given by:
__ __
O4ij = Wijfij = Wij(pijx + qijy + rij) i,j = 1,2
Parameters in this layer are re_ered to as consequent
parameters.

Layer 5:The single node in this layer computes the overall

output as the summation of all incoming signals
__
Figure 3. Fuzzy inference system Out = O5 =2 i=1 2 j=1 Wijfij

where the overall output Out is a linear combination of the

D. ANFIS Layer Description
consequent parameters when the values of the premise
parameters are fixed. It uses sugeno type fuzzy interfernce
A typical ANFIS consists of five layers, which perform systems and Gaussian membership function is used to train
different actions in the ANFIS are detailed below. For the given data set.
simplicity, we have illustrated a system that has two inputs x
and y and one output Z. The rule base, for illustrative IV. WEBBASED ERP MODEL
purposes consists of two if-then rules of the Ta-kagi-Sugeno
type.
Our Education ERP has been designed to cover the in depth
functionalities of any Educational Institute/University/
Group of Institutions, from the perspective of
various users carrying different roles and responsibilities
such as Students, Teachers, Staff, Principal, Management,
Parents, Alumni etc. All the data is managed in a time
sensitive manner along with the rules and policies applicable
at that time, so whenever required, the exact information can
be re-produced as it is. The strength of our Education ERP
increases many fold with the integration of our other ERP
packages like HR, Payroll, Accounts Inventory, Library etc.
However, the entire solution is designed based on a modular
approach that gives flexibility to our clients to choose desired
modules as per their requirements. We have developed an
integrated solution for complete computerization for
Figure 4: ANFIS Layer diagram educational institutions, build on the most futuristic and
highly sophisticated Java environment, denoted as MII -
Layer 1: All the nodes in this layer are adaptive nodes. They Educational Institutes Management System. The solution
generate membership grades of the inputs. has been implemented in many prominent and reputed
The node function is given by: educational institutions of all levels from multi-branch
O1Ai = (x) i = 1; 2 Nursery Schools, Graded Schools to Colleges of the country.
O1Bj = (x) j = 1; 2 Since, this an Integrated, user configurable and dynamic
where x and y are inputs and Ai and Bj are appropriate software solution, it help institutions to get the wide range
membership: functions which can be triangular, trapezoidal, detailed and summarized information of Administrative and
Gaussian functions or other shapes. Academic nature, in different forms required at different
In our study, the Gaussian: MFs has been utilized and three level of the Organizational hierarchy and for other interested
input parameters are: Staff, Team Work and Technical. parties like Students, Parents and other Organizations.
Educational Institute Management System (MII-ERP) is best
Layer 2: The nodes in this layer are fixed nodes which software for schools, Colleges, Institutes, Engineering
multiply: the inputs and send the product out. The outputs of Colleges, management Colleges, medical Colleges, Nursery
this layer are: represented as: with SMS, IVRS, GPRS and web portal. A sample of the data
set showing the independent variables as well as the
Wi=Ai(x)Bj(y) i,j=1,2 dependent variable for the model is given in Table 1.

197 www.erpublication.org
 Comparative study of Institute based ERP based on ANFIS, ANN and MLRA

Data Input
Table 1: Sample Dataset
S. NO. STAFF TEAM WORK TECHNICAL USER
VI. CONCLUSION AND FUTURE ASPECTS
SATISFACTION
1 5 5 5 5
2 1 1 1 1
This study has modeled the ERP Implementation process,
3 3 1.66 3 2
using causal factors Staff, Team Work and Technical as
predictor variables and User Satisfaction as the dependent
4 2.66 2.33 3.33 3
Variable. These factors represent the relevant causal factor
5 1.66 1.33 1.66 2
that impact the success or failure of an ERP implementation
6 3.66 3.66 4.66 4
in term of User satisfaction. We developed/ trained Multiple
7 3.33 3.33 2.33 3
Linear Regression
8 2.33 3 2.33 2
Analysis (MLRA), Artificial Neural Network(ANN) and
9 1.66 2.33 1.33 2
Adaptive Neuro Fuzzy Interference System (ANFIS)
10 2 2 1.66 4 prediction models using part of the dataset(99 responses)
11 2 2.33 2.66 5 prediction models using part of the balance (43 responses).
Of the three techniques ANFIS outperformed ANN and
MLRA in terms RMSE and MAPE. The study established
V. RESULT AND DISCUSSION the efficiency of ANFIS as a good predictor of project risk of
ERP implementation measured to evaluate overall IS
success.
A. Prediction with ANFIS
REFERENCES

[1] New Standish Research Report, Roadmap to the Megap- les, 2009.
http://www.standishgroup.com
[2] D. Robey, J. Ross and M. Boudreau, Learning to Im- plement
Enterprise Systems: An Exploratory Study of the Dialectics of
Change, Journal of Management Informa- tion Systems, Vol. 19, No.
1, 2002, pp. 17-46.
[3] J. Brockner, The Escalation of Commitment to a Failing Course of
Action: Towards Theoretical Progress, Aca- demy of Management
Review, Vol. 17, No. 1, 1992, pp.39-61.
[4] M. Keil, Pulling the Plug: Software Project Management and the
Problem of Project Escalation, MIS Quarterly, Vol. 19, No. 4, 1995,
pp. 421-447.
[5] M. Keil, Why Software Projects Escalatean Empirical Study and
Analysis of Four Theoretical Models, MIS Quarterly, Vol. 24, No. 4,
2000, pp. 631-664.
[6] A. J. Al-Shehab, R. T. Hughes and G. Winstanley, Modelling
Risks in IS/IT Projects through Causal and Cognitive Mapping, The
Electronic Journal of Informa- tion Systems Evaluation, Vol. 8, No. 1,
2005, pp. 1-10.
[7] C. Venugopal and S. Rao, Detecting Project Risks in ERP Projects
Measurement Models for Critical Success Factors and Success of ERP
Implementations, Procced- ings of International Conference on
Advances in Indus- trial Engineering Applications, Chennai, India,
2010.
[8] T. M. Somers and K. Nelson, The Impact of Critical Success Factors
across the Stages of Enterprise Resource Planning Implementations,
Proceedings of the 34th An- nual Hawaii International Conference on
System Sci- ences, Hawaii, 2001, pp. 8016-8025.
[9] W. H. DeLone and E. R. McLean, Information Systems Success: The
Quest for the Dependent Variable, Infor- mation Systems Research,
Vol. 3, No. 1, 1992, pp. 60-95.
[10] W. H. DeLone and E. R. McLean, The Delone and Mclean Model
of Information Systems Successa Ten Year Update, Journal of
Information Systems, Vol. 19, No. 4, 2003, pp. 9-30.
[11] T. H. Davenport, Putting the Enterprise into the Enterprise System,
Harvard Business Review, Vol. 76, No. 4, 1998, pp. 121-131.

B. Comparison between ANN, ANFIS and MLRA

Table 1: Comparison Table
S. NO. ERROR MLRA ANN ANFIS
1 MAPE 8.944 31.38 2.12
2 RMSE 0.9812 1.3833 0.2945

198 www.erpublication.org