Evolution and Search Based Metrics to Improve Defects Prediction

Testing activity is the most widely adopted practice to ensure software quality. Testing effort should be focused on defect prone and critical resources i.e., on resources highly coupled with other entities of the software application.In this paper, we used search based techniquesto define software metrics accounting for the role aclass plays in the class diagram and for its evolutionover time. We applied Chidamber and Kemerer and the newly defined metrics to Rhino, a Java ECMA script interpreter, to predict version 1.6R5 defect prone classes. Preliminary results show that the new metrics favorably compare with traditional object oriented metrics