Predicting change-prone parts in the program is one way to support impact analysis without necessarily reducing the amount of ripple effects. Change-proneness prediction helps to reduce change impact analysis effort by analyzing important entities first which might, in turn, reduce remaining ripple effects. It also provides high chances of identifying valid impacts out of overestimated entities, which might not actually be changed. Several studies showed that coupling measures are correlated to change-prone classes in object-oriented (OO) systems. However, coupling measures are static. Since changes impact propagation could be more dependent on the behavior of the program than program``s structure, the behavioral aspects are needed to be considered for more accurate change-proneness prediction.
In this paper, we present a model-based technique for measuring the behavioral dependency, Behavioral Dependency Measure (BDM), between objects to predict change-prone objects. We measure BDM on UML behavioral models, Sequence Diagrams (SD), to use earlier in software development when the UML design model of a system becomes available. This provides developers and maintainers with early insight into the system by controlling changes. We integrate BDM with weight factors such as execution rate and distance to increase the accuracy of change-proneness prediction. Based on developed measures, we construct Behavioral dependency Matrix (BDX) which helps to predict change-prone objects efficiently. We perform a case study to explain how to perform change-proneness prediction using BDM. Then, changes made on actual implemented codes will show the prediction results are fairly accurate.