You are on page 1of 3

2.

Modeling Methods
2.1 Analysis of Variance (ANOVA)
If p is the number of factors, the ANOVA model is written as follows:
P

 Yi=βo +  k i ,J  , j   i  …………… (3)


j 1

 
2
 i 1wi yi  y i
n
n
1
 R2  1 - , where y = w y …………… (4)
 
2 i i

 i 1wi yi  y i n
n
i 1

 
 2  1 1 R2
R  w w p1 1 …………… (5)
w  y  y 
1 n 2
 MSE = ……………(6)
Wp i 1
i i i

100 n y  y i
 MAPE=  wi i …………… (7)
W i 1 yi

   
2
 y  y  y y 

n
i 1  i i i  i 1

 DW= …………… (8)
 
2

n

w x y
i 1 i i i

SSE
 Cp= + 2p - W…………… (9)

SSE
 AIC=W ln ( ) + 2p…………… (10)
W
SSE
 SBC= W ln ( ) + ln (W) p…………… (11)
W

 PC =
 1  R   W  p  …………… (12)
2

Wp

w  y 1  y  
n 2
 Press = i i i  …………… (13)
i 1
2.2 Multilayer Perceptron (MLP)
Architecture (multilayer perceptron algorithms)
The general architecture for MLP networks is:
Input layer: J0=P units, a0:1,...,a0:j0; with a0:j=xj.
ji 1

ith hidden layer: Ji units, a i:1 ,...,ai:ji ; with ai:k =  i  Ci:k  and Ci:k =  wi: j , k ai 1: j
j 0

where ai 1:0 =1
j1

Output layer: J1=R units, ai:1,……, aI:J ; with a I:k =  I ( c I: k) and c I: k = w


j 0
I: j,k a i-1: j

Where a i-1: 0=1

Note that the pattern index and the bias term of each layer are not counted in the total number of
units for that layer.

Activation Functions
Hyperbolic Tangent
c
  c   tan h  c  = ec  e  c …………… (14)
c

e e
Sigmoid
1
   c = …………… (15)
1  e c
Identity
   c  = c…………… (16)
This is only available for output layer units.
Softmax
e
ck 

   ck  = …………… (17)
 jò  h

This is only available if all output layer units correspond to categorical variables and cross-
entropy error is used.

Error Functions
Sum-of-Squares
M

 ET  w  =  E  w …………… (18)
m
m1

Where
R
1
 Em  w  = ( yr m  aIm:r )2…………… (19)
2 r 1

Cross-Entropy
M

 ET  w  =  Em  w  …………… (20)
m1

Where

 Em  w  =   am
yr m  log  Im:r

 …………… (21)
rò c  yr 

This is only available if all output layer units correspond to categorical variables and the softmax
activation function is used.

You might also like