The urban sprawl dynamics: does a neural network understand the spatial logic better than a cellular automata?



Annexe 2 - Statistical functions for validation of the SANNs

x(p) is the generic output component for the p-th input pattern, and the correspondent
target is
t(p). M is the number of patterns considered for the statistical measure.

Root Mean Squared Error (RMSE), evaluating the squared root of the semi-mean of the squared prediction errors:

1M

RMSE =


2M( x(p )-t ( p ))2

Ï      p=1

Normalized Root Mean Squared Error (NMSE), evaluating the squared root of the mean of the squared prediction
errors, where target and output values was before normalized between 0 and 1:

1M

NMSE = λ--∑

VM z^1

Ï p=1


x(p) - min k {x(k)}
min k {x(k)}- max k {x(k)}


t(p) - min k {t(k)}     J ∣

min k {t(k)}- max k {t(k)}JJ


Real Error (ERR) evaluating the mean of the prediction error:

1M

ERR = -M(χ ( p ) -1 ( p ))

M p =1

Relative Error (ABSERR) evaluating the mean of the absolute prediction errors:

1M

ABSERR = -- ∑ x( p ) -1 ( p ) .

M p =1

Squared R (R2), evaluating the squared of the linear correlation coefficient between target and prediction values:

R2

M ((     ( 1 M   Y

∑ *p)-lMx(k)J

p=1Vk       k k=1     jJ

ʌ 2

(     ( 1 M Yfl

k t ( p )-k Mi∑;( k ) JJJ

U

2

M (     ( 1 M   ll

xp)-l Mx(k)J 4
p=1k        k k=1     JJ    !

2J

M(     (1 M    ll J

t(p )-l -Mt ( k ) J
p=1k       k k=1    JJ J

20



More intriguing information

1. The name is absent
2. The name is absent
3. A Theoretical Growth Model for Ireland
4. The name is absent
5. The ultimate determinants of central bank independence
6. The value-added of primary schools: what is it really measuring?
7. The name is absent
8. Wirkung einer Feiertagsbereinigung des Länderfinanzausgleichs: eine empirische Analyse des deutschen Finanzausgleichs
9. Declining Discount Rates: Evidence from the UK
10. Dynamic Explanations of Industry Structure and Performance