Enhanced Generalized Regression Neural Net for Breast Cancer Detection

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 16, No. 1, January 2018
Enhanced Generalized Regression Neural Net for
Breast Cancer detection
Samira Babaei Ghalejoughi1
Department of Electronic Engineering.
Tabriz Branch, Islamic Azad University.
Tabriz, Iran.
1
stu.samira.babaei@iaut.ac.ir
Nasser Lotfivand*2
Department of Electronic Engineering.
Tabriz Branch, Islamic Azad University.
Tabriz, Iran.
*2
lotfivand@iaut.ac.ir
Abstract—In this paper we represent a modified Generalized
Regression Artificial Neural net that can recognize all breast
cancer of Wisconsin Diagnostic Breast Cancer and Wisconsin
Prognostic Breast Cancer correctly. In this method the modified
Neural Net trained with 50% of data & 50% for test. But the
result is the ability of classify with 100% accuracy. The all 50%
train & test data chosen randomly.
This method is based on the fact that calculation in float numbers
will remove accuracy. By reducing the number of calculation the
accuracy of result increase significantly.
Keywords-neural network, Generalized regression neural
network(GRNN), absolute distance.
I. INTRODUCTION (HEADING 1)
Pattern classification problems are important application
areas of neural networks used as learning systems [1],[2],[3].
Multilayer Perceptrons (MLP), radial basis functions (RBF),
probabilistic neural networks (PNN), self-organization maps
(SOM), cellular neural networks (CNN), recurrent neural
networks and conic section function neural network (CSFNN)
are some of these neural networks. In addition to classification
problems, function approximation problems are also solved
with neural networks. Generalized regression neural network
(GRNN) is one of the most popular neural network, used for
function approximation. GRNN and PNN are kinds of radial
basis function neural networks (RBF–NN) with one pass
learning [1]. However they are similar; PNN is used for
classification where GRNN is used for continuous function
approximation [4]. But in this paper we use GRNN for
recognizing.
II. RELATED WORK
In Back Propagation neural net the neurons trained with
gradient descent algorithm the final weight change is (1)
   . . . . . .( 2 1     pq k p q q q k q k q k p jW η * * α* T φ * φ * φ * φ
 
   
. .
1
. .
. . , . .
2
1 1

 
  
    
   

r
q q k q k
q
hp j p q
q k p j q k p j p j h
* α* T φ * φ *
W η *
φ * φ *W * α* φ * φ * x

The Wp,q is the weights between second & third layer &
Wh,p is the weights between first & second layer.
T he structure of GRNN Neural net shown in figure(1). The
learning algorithm shown in (2).

N
k kK 1
N
kK 1
y K(x,x )
Y(x)
K(x,x )





 
The pros of GRNN is that it can learn in one train. The cons
is that it need to save all training data & in some case this need
big memory. The K(x,Xk) is radial base function & the formula
for K(x,Xk) is shown in (3). Yk is the prediction value for Xk.
Y(x) is the prediction value for x.
where kd is the squared Euclidean distance between the
training samples kX and the input x In huge data the error
increase because of calculating the Euclidean distance.

2
kd /2 T
k k k kK(x,x ) e , d (x x ) (x x )
     
172 https://sites.google.com/site/ijcsis/
ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 16, No. 1, January 2018
III. PROPOSED WORK
In the proposed method instead of calculation of Euclidean
distance we use absolute distance between samples Xk & input
X.
IV. RESULT
To test the simulation we use the Wisconsin Diagnostic
Breast Cancer and Wisconsin Prognostic Breast Cancer. First
we use 50% of data to train & 50% for test. Then we go one
more step & use 40% for train & 60% for test. The result
shown in table 1 & table 2.
The result of GRNN enhanced by changing the Euclidean
distance to Absolute distance.
TABLE I. RESULT OF SIMULATION FOR 50% TRAIN & 50% TEST.
dataset WDBC WPBC Comment
Number of
instances
569 198
Train Percent 50% 50%
Test Percent 50% 50%
Back propagation 95.08% 95.08% Hidden = 10
Linear SVM 78.12% 79.13%
Euclidean distance
GRNN
94.16% 96.14% .01σ
Absolute distance
GRNN
100% 100% .01σ
TABLE II. RESULT OF SIMULATION FOR 40% TRAIN & 60% TEST.
dataset WDBC WPBC Comment
Number of
instances
569 198
Train Percent 40% 40%
Test Percent 60% 60%
Back propagation 90.02% 91.49% Hidden = 10
Linear SVM 78.44% 79.13%
Euclidean distance
GRNN
93.84% 93.84% .01σ
Absolute distance
GRNN
100% 100% .01σ
V . CONCLUSION
In real world we need big data. If we use many calculation
the accuracy of computers become low. By reducing the
number of calculation we improve the accuracy.
REFERENCES
[1] Al-Daoud, E. (2009). A comparison between three neural network
models for classification problems. Journal of Artificial Intelligence, 2,
56–64.
[2] Bartlett, P. L. (1998). The sample complexity of pattern classification
with neural networks: the size of the weights is more important than the
size of the network. IEEE Transactions on Information Theory, 44(2),
525–536.
[3] Specht, D. F. (1990). Probabilistic neural networks. Neural Networks, 3,
109–118.
[4] Specht, D. F. (1991). A general regression neural network. IEEE
Transactions on Neural Networks, 2(6), 568–576.
Figure(1) The GRNN Neural Network
173 https://sites.google.com/site/ijcsis/
ISSN 1947-5500

Enhanced Generalized Regression Neural Net for Breast Cancer Detection

More Related Content

What's hot (18)

Similar to Enhanced Generalized Regression Neural Net for Breast Cancer Detection (20)

Recently uploaded (20)

Enhanced Generalized Regression Neural Net for Breast Cancer Detection