Neural Network Modelling of Constrained Spatial Interaction Flows



of two overall performance measures. The first performance measure are the average
relative variances, ARV
( M3 ), a normalised mean squared error metric that is widely
utilised in the neural network community:

ARV ( M3 ) =


∑   (yu3 - 1Ωll (xu3, w))2

( xu 3, yu 3 ) M3


∑  (y3 - yu3 )2

( xu 3, yu 3 ) M3


-31-  ∑  (yu3 - ⅛ (xu3, w))2(32)

( U ? - u3 u3 ,, '                            7

3 ( x , y I M3


where ( xu 3, yu 3 j denotes the u 3-th pattern of the testing set M3, yu 3the average over
the
U3 = 248 desired values. The averaging, that is the division by U3 [the number of
patterns in
M3], makes ARV independent of the size of M3 . The division by the
estimated
σ2 of the data removes the dependence on the dynamic range of the data.
This implies that if the estimated mean of the observed data would be taken as
predictor, ARV would equal to one (Weigend, Rumelhart and Hubermann 1991). The
statistic has a lower limit of zero indicating perfectly accurate predictions and an upper
limit that is in practice one.8

The second performance measure is the standardised root mean square error (SRMSE)
that is widely utilised by spatial analysts (see Fotheringham and Knudsen 1987):

1

. . 1 Γ , ... .      .... ..2       13

SRMSE(M.) = —   X (yu3 - 1 l (xu3, w)) /U3

(33)


y ( xu 3, yu 3 ) M3

This statistic - closely related to the ARV-measure - has a lower limit of zero
indicating perfectly accurate predictions and an upper limit that is variable9 and
depends on the distribution of the
yu3 .

5.2 The Data

The testbed for the benchmark comparisons uses interregional telecommunication
traffic data for Austria. From three Austrian data sources - a (32, 32)-interregional

21



More intriguing information

1. Density Estimation and Combination under Model Ambiguity
2. The Works of the Right Honourable Edmund Burke
3. A novel selective 11b-hydroxysteroid dehydrogenase type 1 inhibitor prevents human adipogenesis
4. The name is absent
5. WP 92 - An overview of women's work and employment in Azerbaijan
6. EU Preferential Partners in Search of New Policy Strategies for Agriculture: The Case of Citrus Sector in Trinidad and Tobago
7. APPLICATIONS OF DUALITY THEORY TO AGRICULTURE
8. SLA RESEARCH ON SELF-DIRECTION: THEORETICAL AND PRACTICAL ISSUES
9. Non-causality in Bivariate Binary Panel Data
10. The name is absent
11. A Study of Adult 'Non-Singers' In Newfoundland
12. Tobacco and Alcohol: Complements or Substitutes? - A Statistical Guinea Pig Approach
13. Understanding the (relative) fall and rise of construction wages
14. Placenta ingestion by rats enhances y- and n-opioid antinociception, but suppresses A-opioid antinociception
15. Strategic Investment and Market Integration
16. The name is absent
17. The Response of Ethiopian Grain Markets to Liberalization
18. On the Relation between Robust and Bayesian Decision Making
19. Corporate Taxation and Multinational Activity
20. AGRIBUSINESS EXECUTIVE EDUCATION AND KNOWLEDGE EXCHANGE: NEW MECHANISMS OF KNOWLEDGE MANAGEMENT INVOLVING THE UNIVERSITY, PRIVATE FIRM STAKEHOLDERS AND PUBLIC SECTOR