Jacoco-Coverage Based Statistical Approach For Ranking And Selecting Key Classes In Object-Oriented Software

Ranking and selecting the most fundamental classes in a software is a critical task for recognizing an inexperienced system. There are several approaches to detect these classes. This paper proposes a statistical-rank based approach that addresses an untouched area of testing coverage usefulness. Testing coverage information is used as new method to select the key classes in an object-oriented software. The proposed approach comprises three statistical phases including ranking the software classes, selecting the highly coverage classes, and selecting the key classes. Experiments are conducted on 23 successive releases of large opensource software system, Apache Lucene and three different projects are investigated to validate the proposed approach. A multiple linear regression model has been applied both to evaluate the proposed approach and to show the relationship between testing coverage of classes and their structural properties. The most significant predictors for testing coverage are complexity, coupling, lack of cohesion, size, and inheritance which constitute properties of the software classes. The testing coverage percentage of the key classes ranged between 72.7% - 82.6%. Our approach aims to help software engineers to assess software design and better achieve testing objectives.