Combining probabilistic models for explanatory productivity estimation

In this paper Association Rules (AR) and Classification and Regression Trees (CART) are combined in order to deliver an effective conceptual estimation framework. AR descriptive nature is exploited by identifying logical associations between project attributes and the required effort for the development of the project. CART method on the other hand has the benefit of acquiring general knowledge from specific examples of projects and is able to provide estimates for all possible projects. The particular methods have the ability of learning and modelling associations in data and hence they can be used to describe complex relationships in software cost data sets that are not immediately apparent. Potential benefits of combining these probabilistic methods involve the ability of the final model to reveal the way in which particular attributes can increase or decrease productivity and the fact that such assumptions vary among different ranges of productivity values. Experimental results on two data sets indicate efficient overall performance of the suggested integrated method.