Predicting Costly Maintenance Packages Using Package Cohesion

Abstract

Poorly designed software systems leads to difficult and costly maintenance activities. Object-oriented software metrics try to reflect the quality of the system design. Package cohesion metrics are found to be related to software maintainability. In our previous work, we developed a new package cohesion metric based on Martin's well-known and well-accepted package design principles. This metric was found to have a correlation with software maintainability and software testability. In this paper, we continue this empirical investigation by predicting costly maintenance packages using package cohesion. The prediction models created by logistic regression analysis show that costly packages can be predicted using package cohesion as early as the design phase of the software development lifecycle. This early prediction helps software engineers to discover any weak designs of the system and then these packages can be redesigned or at least carefully tested and documented, which contribute to considerable reductions in maintenance costs. The results demonstrate that the new package cohesion metric is promising and can help in locating low maintainable software packages.