Modelling software reliability using hybrid Bayesian networks

In this research, we analyse the problem of predicting software reliability from AI perspective.We observe that existing models are built based on expert knowledge including defining aset of metrics through surveys and causal relationships. We overcometheir limitations by introducing new data collection, model construction and inferencemethodology. We propose a Hybrid Bayesian network that would estimate reliabilityof consecutive releases of software projects before a release decision, in terms of theirresidual (post-release) defects. We form this hybrid model by incorporating quantitativefactors of development and testing processes into qualitative factors of requirementsspecification and documentation process without the need for any transformation.As quantitative factors, we select popularly used product, in-process and people metricsas well as introduce new ones depending on the availability of local data in the organizations.We also identify qualitative factors representing requirements specificationprocess via surveys with development teams. Dependencies between software metricsand defects are determined according to correlation and independence tests andgraphical dependence analysis with chi-plots. We utilize a Monte Carlo technique toapproximate joint probability distribution of the model over conditionals by inferringunknown distribution parameters. Empirical analyses on two industrial datasets showthat (i) Hybrid Bayesian networks are capable of estimating reliability in terms ofresidual defects, (ii) proposed way of defining causal relationships, chi-plots, decreaseserror rates signicantly, (iii) expert judgement-based models may not achieve as goodprediction performances as statistical models, (iv) local data are so valuable and representativeas expert knowledge in software organizations that they should be usedprimarily and strengthened with expert knowledge in predicting software reliability.