Measuring Software Reliability: A Trend Using Machine Learning Techniques - Complex System Modelling and Control Through Intelligent Soft Computations - page 804

Fig. 5 a versus t with error
Fig. 6 a versus t without
error
set is taken as three quarters of the given data set and the remaining one quarter is
used for the testing purpose. In the training process, the whole data set is nor-
malized using Z-score normalization and the standard deviation of the normalized
values is taken as the standard deviation, r, for the Gaussian kernel function which
is used for the mapping of the input data elements into a higher-dimensional space,
and the value obtained is 1.0545. The value of C is then tuned by comparing the
predicted and observed values of the failure data of each week in the training data
set such that minimum loss occurs. The same function is used to choose appropriate
values for the Lagrange multipliers, i, i*, used to solve the optimization problem,
which are then used for the rest of the training process. Using these values, the
value of the bias term, b, is tuned in a similar way as that of C, that is, by comparing
the predicted and observed values of the data elements in the training data set so