Information Technology Reference
In-Depth Information
∑
,,
,,
1
,
(7)
∑
,,
,,
,
where
x
l,T,temp
is the attribute net temporal variation of the target case
T
with respect to
attribute
l, l=k,s.
and
x
l,C,temp
is the attribute net temporal variation of case
C
, x
l,T,spatial
is the attribute net spatial variation of the target case
T
with respect to attribute
l
and
x
l,C,spatial
is attribute net spatial variation of case
C
. The normalisation factor is given
by
norm
l,,temp
and
norm
l,spatial
.
•
Aggregate similarity
The aggregate similarity
S
All
is given by the average of the sum of the individual simi-
larity values between the target case and a case from the case base.
3
(8)
4
Validation of the CBR Methodology
The aim of this study is to investigate the feasibility of using CBR to predict the like-
lihood of an accident happening under certain traffic conditions. In order to be able to
do this, we need to ascertain that the flow data contains certain properties that are
indicative of accidents. In other words, there have to be flow data properties that al-
low us to differentiate between accident flow data and non-accident flow data. The
basic premise of CBR is that similar cases have similar solutions. In our problem, we
can interpret the premise as similar cases having similar outcomes in terms of acci-
dents. The question is if given a target case of flow data attributes, does the most
similar case to the target case retrieved from the case base, have an outcome that can
be expected for the target case as well? Currently, the only input data available to the
system is the record of accidents and the traffic flow data for both accidents and non-
accidents. However, it is still uncertain if by evaluating exclusively the flow data
online using CBR it is possible to predict accidents.
The success rate of a CBR system can be quantified in terms of the number of suc-
cessful retrievals from a sample of test cases (i.e., cases in which the solution to be
determined is known). Retrieval is deemed successful if the solution or outcome of
the retrieved case, with the highest similarity to the target case, is considered correct.
As we know the outcome of the test cases, i.e., which cases constitute accident cases
and which cases constitute non-accident cases, a successful retrieval is Outcome (
C
T
)
= Outcome (
C
C
), i.e.:
