Measuring Software Reliability: A Trend Using Machine Learning Techniques - Complex System Modelling and Control Through Intelligent Soft Computations - page 809

Fig. 13 Emperical data set emperical

6 Conclusion

Machine learning has become phenomenal over statistical estimation techniques by
transforming various industrial and scienti fields for over the last decade.

Equipped with the capability to handle large, complex and categorical data; cope
with non-existent or inconsistent data values; and reason on the basis of previous
data, machine learning has offered tremendously improved results. This project
shows how support vector regression, a machine learning based method when
applied to software reliability domain gives better estimation and prediction in
comparison to its statistical counterpart, Goel-Okumoto modeling. However, per-
formance of SVR is limited by the heuristic basis used in parameter estimation. The
range of values used for tuning the learning parameters must be carefully chosen for
desired accuracy. Fully automating this aspect is still an associated challenge which
needs to be handled in future. Another area of future study concerns the compu-
tation of Lagrange Multipliers for support vectors. As their values can lie anywhere
over the range [0, C], it becomes dif

c

cult to determine their exact values. Hence,
instead of choosing random values, appropriate heuristic function must be designed
for this purpose to make the method more deterministic.