Академический Документы
Профессиональный Документы
Культура Документы
KDSI
Many different models of neural networks have been
proposed [13]. They may be grouped in two major RELY
categories. First, feed-forward networks, in which no loops Effort
ACAP
in the network path occur. Second, feedback networks that -
have recursive loops. The feed-forward multi-layer AEXP
Perceptron with the Backpropagation learning algorithm are
the most commonly used in the cost estimation field. In
these nets, neurons are arranged in layers and there are only
connections between neurons in one layer to the following. SCED
Figure 1 illustrates a possible network architecture
configured for software development effort estimation. The Figure 1: A neural networkarchitecture
for software development effort
network generates output (effort) by propagating the initial
inputs (cost drivers or project attributes) through subsequent
As mentioned earlier, the use of a neural network
layers of processing elements to the final output layer. Each
approach to estimate software development effort requires
neuron in the network computes a nonlinear function of its
certain decisions and choices about the architecture,
inputs and passes the resultant value along its output. The
learning algorithm and the activation functions. In our case,
favored activation function is the Sigmoid function:
the method, that we use to generate the if-then fuzzy rules
I
m=-
I
(1) from the neural network, requires that the architecture must
be a three-layer Perceptron with the activation functions of
Before the network is ready to make estimates for new the hidden layer and the output layer are the sigmoid and the
software projects, it is trained by a set of combinations of identity functions, respectively. Our neural network has 13
inputs and outputs that are known as the training data. Our inputs (COCOMO cost drivers) and one output
case study consists of estimating the software development (development effort). All the inputs as well as the output of
effort by using the neural networks approach on the the network are numeric. The inputs are normalized to speed
COCOM0’8 1 dataset. The COCOM0’8 1 dataset contains up the training process of the network [14]. The network is
63 software projects [2,3,4]. Each project is described by 17 trained by iterating through the training data many times.
attributes: the software size measured in KDSI (Kilo The learning algorithm used is Backpropagation with
Delivered Sources Instructions), the project mode defined as teaching rate and maximum error are equals to 0,03 and lo’,
either ‘organic’, ‘semi -detached’ or ‘embedded’, and the respectively.
remaining 15 attributes are measured on a scale composed
of six linguistic values: ‘very low’, ‘low’, ‘nominal’, ‘high’,
-
TABLE 3: TWO EXAn4PLE S OF THE OBTAWED FU ZZY RULES
117.57 30.14
(a) (b)
Figure 2: (a) Fuzzy set associatedto the qualification ‘approximately greater than 117,57’.
@)Fuzzy set associated to the qualification ‘ approximately not greater than 30,14’
r I r I r 1
5 D/P
10 55 100 550 loo0
Figure 3: Membership functions of fuzzy sets defined for the DATA cost driver (61
I I I I I
-L.
0,5 1 2 4 8 12 36 921,91
Figure 4: Example where the value used for the LEXP attribute is out of ran ge values
6. CONCLUSION AND FUTURE WORK [6] A. Idri, L. Kjiri, and A. Abran, “COCOMO Cost Model Using Fuzzy
Logic”, 7* International Conference on Fu zzy Theoly & TechnologW
Atlantic City, NJ, February, 2000. pp. 219223
In this paper, we have studied one of the most important [7] A. Idri, and A. Abran, “Towards A Fuzzy Logic Based Measures For
limitations of neural networks, that is understanding why Software Project Similarity”, Sixth Maghrebim Conferenceon Computer
an neural network makes a particular decision is a very Sciences, Fes, Momoco, November, 2000. pp. 9- 18
[8]A. Idri, and A. Abran, “A Fuzzy Logic Based Measures For Software
difficult task. Our study is intended for the cost estimation Project Similarity: Validation and Possible Improvements”, 7‘h
field. The neural network that we have used to predict the Intemational Symposium on Software Metrics, lEEE computer socieQ,4-
software development effort is the Backpropagation three- 6 April, England, 2001. pp. 85 -96
layer Perceptron with the sigmoid function in the hidden [9] A. Idri, and A. Abran, “Evaluating Software Project Similarity by
using Linguistic Quantifier Guided Aggregations”, b IFSA World
units and the identity function in the output unit. We have Congresd20 th NAFIPS Intemational Conference, 25 -28 July, Vancouver,
used the entire COCOM0’81 dataset to train and to test 200 1. pp. 41 6-421
the network. It is observed that the obtained accuracy of [IO] A. Idri, A. Abran, T. M. Khoshgoftaar, “Fuzzy Analogy: A new
the network is acceptable. Approach for Software Cost Estimation”, 11 Intemational Workshop in
Software Measurements,28-29 August, Montreal,2001,pp: 93-101
After training and testing the network, we have applied [ I l l A. Idri, T. M. Khoshgoftaar, A. Abran, , “Btimating Software
the Benitez’s method to extract the if-then fuzzy rules Project Effort by Analogy based on Linguistic values”, To be presented
from this network. These fuzzy rules express the in 9 IEEE Intemational Software Metrics Symposium”, 47 Ottawa,
information encoded in the architecture of the network. Canada, 2002
The interpretation of each fuzzy rule is determined by [12] J. S. Jang, C. T. Sun, ‘Functional equivalence between radial basis
fimction networks and f i z z y inference systems’, IEEE Transaction m
analyzing its premise and its output. Our case study shows Neural Networks, Vol. 4,1992, pp. 156-158
that we can explain the meaning of the output and the [ 131 M Jorgersen, ‘Experience with Accuracy of Software Maintenance
propositions composing the premise of each fuzzy rule. Task Effort Prediction Models’, IEEE Transaction on Software
Engineering, Vol. 21(8), 1995, pp. 674-681
However, the b r operator seems to be inappropriate to [I41 A. Lapedes, Farber R., ‘Nonlinear signal prediction using neural
combine the effects of the various fuzzy propositions on networks’, Prediction and System modeling’, Los Alamos National
the output of the fuzzy rule in software cost estimation. Laboratory, Tech. Report, LA-UR-87-2662,1987
Consequently, we are currently exploring the use of other [15] R. P. Lippman, , ‘An Introduction to computing with neural nets’,
IEEE ASSP Mag, vol. 4, pp.4-22, 1987
methods in order to extract more understandable fuzzy [I61 B. Samson, Ellison D., Dugard P, ‘Software Cost Estimation using
rules from the type of neural network used in this case an Albus Perceptron’, @ International COCOMO Estimation meeting,
study. Pittsburgh, 1993
[ 171 C. Schofield, ‘Non-Algorithmic Effort Estimation Techniques’,
Tech. Report TR9801, March, 1998
7. BIBLIOGRAPHY [ 181 C. Serluca, ‘An Investigation into Software Effort Estimation using
a Backpropagation Neural Network, MSc. Thesis, Bournemouth
[I] J. M. Benitez, J.L. Castro, I. Requena, ’Are Artificial Neural University, 1995
Networks Black Boxes?’, IEEE Transaction on Neural Networks, Vol. 8, [I91 K. Srinivasan, Fisher D, ‘Machine Leaming Approaches to
NO. 5, September, 1997, pp. 1156-1 164 Estimating Software Development Effort’, IEEE Transactionon Software
[2] B.W. Boehm, Software Engineen’ng Economics, PrenticsHall, 1981. Engineering, Vol. 21, No. 2, February, 1995, pp. 126-136
[3] B.W. Boehm, and d.,“Cost Models for Future Software Life Cycle [20] A. R. Verkatachalam, ‘Software Cost Estimation using Artificial
Processes: COCOMO 2.0”, Annals of Software Engineen‘ng on Sof”? Neural Networks, Intemational Joint Conference on Neural Networks,
Process and Product Measurement, Amsterdam, 1995. Nogoya, IEEE, 1993
[ 4 ] D.S. Chulani, “Incorporating Bayesian Analysis to Improve the [21] G. Wittig, G. Finnie, ‘Estimating Software Development Effort with
Accuracy of COCOMO I1 and Its Quality Model Extension”, Ph.D. connectionist Models’, Information and Software Technology, vol. 39,
Qualifying Exam Report, USC, February, 1998. 1997, pp.469476
[5] R.T Hughes, ‘An Evaluation of Machine Leaming Techniques for [22] L.A. Zadeh, “Fuzzy Set”, Information and Control, Vol. 8, 1965,
Software Effort Estimation’, University of Brighton, 1996 pp. 338-35