Bayesian Belief Networks as a Software Productivity Estimation Tool

Defining the required productivity in order to complete successfully and within time and budget constraints a software development project is actually a reasoning problem that should be modelled under uncertainty. The contribution of this paper is the analysis of the applicability of probabilistic reasoning approaches, in particular Bayesian Belief Networks (BBN), to this problem. BBNs are capable of discovering the dependencies and independencies among the attributes of a project and defining the direct impact of some of them on productivity. Uncertainty is depicted through the use of estimate intervals and probabilities: the estimation is actually an interval within which the productivity of a project is likely to fall in, with a certain probability, considering both an optimistic and a pessimistic situation. The use of predefined intervals is another important feature of the method, allowing the control of the estimation process and the generation of meaningful intervals, appealing and understood by software managers. The ability of the method to classify correctly the rest of the attributes in one of their discrete values is also tested, paying further attention on the software development mode. The method is applied and evaluated on the widely known COCOMO81 dataset. The evaluation shows that BBN is a promising method whose results can be confirmed intuitively. BBN are easily interpreted, allow flexibility in the estimation, can support expert judgment and create models considering all the information that lay in a dataset by including all productivity factors in the final model.