Information Technology Reference
In-Depth Information
4.3.3 Parameter Calculation
2
, C, and
As there are no general rules to identify the free parameters
, the
optimum values are determined by the approach as follows. First, the values of any
two parameters are
r
fixed in any constants, and then adjusting the value of the third
parameter until minimum forecasting error occurs. Second, set the value of the
parameter determined in previous step; the value of another parameter is also
xed,
and similarly, the value of the third parameter is adjusted until minimum forecasting
error appears. Third, repeat both the previous steps until these three parameters have
been identi
ed.
4.3.4 Reliability Measurement
Two commonly used measures of software reliability are goodness-of-
t and next-
step predictability. The goodness-of-
fitting a curve corre-
sponding to the proposed approach to all the data points in the training data set. The
deviation between the observed and the
t is determined by
fitted values of the number of cumulative
failures per week is then evaluated. The next step predictability is determined by
feeding the unknown data set
to the training model. The input values
x
i
k
;
x
i
are used to predict the value of xi
i
þ
1
where k denotes the number
of input nodes considered. Then the predicted and actual values of the cumulative
number of failures per week are compared. Relative error (RE) is used to represent
the results of the above mentioned measures. The relative error (RE) is de
x
i
k
þ
1
; ...;
ned as
x
i
0
x
i
x
i
RE
¼j
j
ð
9
Þ
where x
i
denotes the predicted value of the number of cumulative failures per week,
and x
i
denotes the actual value of the number of cumulative failures per week.
5 Results and Analysis
After accomplishing the algorithm is previous section, we obtain various test
simulation results by using MATLAB,
1
Sparse Modelling Software; with R
package (refer Appendix).
1
Search WWH ::
Custom Search