Measuring Software Reliability: A Trend Using Machine Learning Techniques - Complex System Modelling and Control Through Intelligent Soft Computations - page 800

4.3.2 Input and Output

The given data set is divided into two sets: training set and testing set. The training
set is used to build the model and the testing set is used to evaluate the model. The
training data set is fed into the SVR model and the parameters that lead to the best
accuracies are selected. We use the recent k data elements seen so far to assess the
software reliability. The training model can re ect the mapping of input and output

of this process by learning a set of training data pairs

fl

ð x i ; x i þ 1 Þ where the observed

data is within the sliding window of size k. The input ð x i k ; x i k þ 1 ; ...; x i 1 Þ is fed

into the SVR model and the corresponding target value is .

After the training process, the SVR model has learnt the inherent correspondence of
the software failure process between these two vectors. Therefore, on giving an
input value xi, i , the predicted value of xi, i þ 1 can be obtained. When each new data
element is arrived, the training and prediction processes of SVR model are per-
formed alternately. For example, if the (i + 1)th data element is arrived, the model
could be trained again with new input vector

ð x i k þ 1 ; x i k þ 2 ; ...; x i Þ

ð x i k þ 1 ; x i k þ 2 ; ...; x i Þ and target

vector and then the trained model can be used to predict

the value of xiþ2. i þ 2 . In this approach, all available failure data are not used, instead
only the data elements in the sliding window are used. This is because the early
failure behavior may have less impact on the later failure process (Figs. 1 and 2 ).

ð x i k þ 2 ; x i k þ 3 ; ...; x i þ 1 Þ

Fig. 1 SVR training process

Fig. 2 SVR prediction process