Software metrics selection model for predicting maintainability of object-oriented software using genetic algorithms

Software development life cycle maintenance has been advocated as the critical part that consumes more time and resources. To understand the magnitude of the task to maintain the software product, software metrics have been used to make quantification based on their respective software features. To predict software maintainace, the proper metrics need to be selected to avoid the duplication or the outlying of the potential metrics. This is because on one hand, the individual metrics deals with only a single feature of the object-oriented systems, while on the other hand; the suites either contain duplicate metrics of the same goal or lack some important metrics that match the common attributes in the software products. The latest effort to solve this selection problem is the development of the metrics selection model that uses genetic algorithm (GA). However, the process failed to state clearly the encoding strategy in its initial stage. This thesis clarifies the issue using the objective method to develop the GA metric selection model for predicting the maintainability of object-oriented systems. The study proposes the use of software metric thresholds in the classification process during the GA representation. The software metric thresholds were used as indication for identifying unsafe design in software engineering. To evaluate this technique, an experiment was conducted on two geospatial systems developed using Java programming language where the Chidamber and Kemerer (CK) metrics were used. The proposed technique was also compared to the ranking results from the experts. The comparison results obtained when compared with those of Principal Component Analysis and the complete software metric suite were very promising. Moreover, the three techniques show significant differences in both treatments when compared using analysis of variance