Generalized Logit Regression-Based Software Reliability Modeling with Metrics Data

Generalized Cox Proportional Hazards Regression-Based Software Reliability Modeling with Metrics Data

“…Looking at these results, it can be seen that the selection of software test metrics influences rather the goodness-of-fit performance. In the earlier work [15], the authors show that the logistic regression-based SRMs with metrics M 1 and M 3 give the best performance in terms of both AIC and MSE, and that that the cumulative number of TPRs is not always significant in DS2. However, this is not true for our generalized Cox proportional hazard regressionbased SRMs.…”

“…(15). Here we compare our generalized Cox proportional hazards regression-based SRMs with those ones from both of goodness-of-fit and predictive performance tests.…”

“…In Kuwa and Dohi [15], eight logistic regression-based SRMs are proposed and compared with the basic Cox proportional hazards regression-based SRM with constant baseline hazard rate in Eq. (15).…”

“…The conclusion there is that the model superiority between the Cox proportional hazards regression-based SRMs and the logistic regressionbased SRMs strongly depends on the kind of data, so that the former may be superior to the latter in some case, and vice versa. Recently, Kuwa and Dohi [15] introduce several kinds of logistic function and develop generalized logistic regression-based SRMs within an NHPP framework. In their numerical comparison, it is shown that the generalized logistic regression-based SRMs often outperform the Cox proportional hazards regression-based SRM with constant baseline hazard rate in terms of the goodness-of-fit performance based on the Akaike information criterion.…”

“…Also, we predict the number of software faults detected in future and also evaluate the predictive performance, where the prediction likelihood and the prediction mean squares error are used. Through the validation experiments, we compare our generalized Cox proportional hazards regression-based SRMs with the existing ones in [25], [27] and the several logistic regression-based SRMs in [15] on the goodness-of-fit and predictive performances. Based on these results, we assess the quantitative software reliability with the best fitted SRM to the past observation.…”