International Journal of Advance Research and Innovation

a 
= β
0
 + β
1
 . V + 
β
2
 . f + β
3
 . a + β
4
 . V
2
 + β
5
 . f
2 
+ β
6 
. a
2 
+
 
β
7 
. V. f + β
8 
. V . 
a + β
9
 . a . f
The multiple regression models are tested by aiding the 
analysis of variance (ANOVA). The data have been used to 
build the multiple regression model. The coefficients β
0
, β
1
, 
β
2
. . . β
9 
are estimated with the least square method using 
MINITAB 14. Multilayer perception (MLP) architecture 
with back-propagation algorithm having two different 
variants is used in neural network. The performances of 
multiple regression and neural network-based models are 
compared by means of statistical methods. In this study, 
ANN structure is used for modeling and predicting surface 
roughness in turning operations. This fully connected 
hierarchical network structure has an input layer, a hidden 
layer, and an output layer. The back-propagation learning 
algorithms such as scaled conjugate gradient (SCG) and 
Levenberg–Marquardt (LM) were used to update the 
parameters in feed forward single hidden layers. 
The cutting speed (V), feed (f), and depth of cut (d) 
were considered as the process parameters. The input layers 
of the neural network consist of three neurons whereas the 
output layer had a single neuron that represents the 
predicted value of surface roughness. The logsig processing 
function and single hidden layer had been used. A trial and 
error scheme had been used to determine the appropriate 
number of hidden neurons. The number of hidden neurons 
was determined as four and five neurons. The maximum 

---

Volume 2, Issue 3 (2014) 676-683 
ISSN 2347 - 3258
International Journal of Advance Research and Innovation 
681 
IJARI 
number of epochs and the learning rate value for each run 
were selected as 10,000 and 0.9, respectively.