Information Technology Reference
In-Depth Information
4.2 Identi
cation of the Elements of the Connection Matrix
Using BEA
In our second part of our research (Buruzs et al.
2013a
), the model uses two
different sets of input data. The sources of these two sets are different; one set is
based on observations that may be considered more or less objective; observations
on the trend of the studied factors in the time period from the 1980s till the 2010s. It
is obvious that measuring the mutual in
uence of various factors within a complex
phenomenon, like waste management is not easy. Nevertheless, it might be
assumed that the time series published in the related literature (Demirbas
2011
; den
Boer and Lager
2007
; Graymore et al.
2008
; Langa et al.
2006
; Morrissey and
Browne
2004
; Wilson et al.
2001
; van de Klundert and Anschutz
1999
; Thorneloe
et al.
1999
) is based on a consensus concerning the interrelationship of the concepts
playing a determinative role in the procedure of waste management, thus these
values are widely supported by independent observations and manually calculated
partial models. In this research, the following data will be considered
fl
'
objective
'
,
even though they are not obtained by
of some automatic
machinery, but by the observation and evaluation of humans involved in the
management of the procedure. It must be clearly understood that our learning model
is based on these
'
measurements
'
data and therefore it makes it unnecessary to consult
continuously the experts in order to obtain again and again up-to-date but entirely
subjective data.
Nevertheless, in order to speed up the learning procedure, and to some extent,
out of scienti
'
objective
'
c curiosity, we used the data collected from the above mentioned
survey. It must be stressed that results of these questionnaires (which were com-
pared, and the medium values selected for each matrix element as the
'
typical
'
subjective values
uence) were used only as initial values for the
learning procedure, under the assumption that starting with more or less realistic
values would speed up the convergence of the matrix to the stable
of the given in
fl
'
values. It turned out during the optimization that the convergence speed is quite
high with randomly generated start population as well, thus prudent composition of
the bacteria in the
'
objective
first generation was not an important issue. It is nevertheless
interesting to compare the
'
subjective
'
mutual in
fl
uence values obtained from the
questionnaires and the
matrix obtained from the time series observed
starting with the data from the 1980s. On the basis of the gathered data we con-
structed the initial draft of the connection matrix, including identi
'
objective
'
cation of concept
nodes and their mutual relationships represented by the graph edges.
Simulation in this context consisted of computing the states of the system
described by the state vector over a number of successive iterations. In every
iteration cycle the state vector speci
es the current values of all factors (the nodes)
in a particular moment. The values of the given states (nodes) are obtained from the
preceding iteration values of all the nodes, which exert in
uence on the given node
through cause-effect relationship. The transformation function is used to con
fl
ne the
weighted sum to the range set to [0, 1]. This normalization hinders the absolute
Search WWH ::
Custom Search