An Empirical Study on Object-OrientedMetrics

Author(s): Mei-Huei Tang, Ming-Hung Kao, and Mei-Hwa Chen
Venue: Software Metrics Symposium, Proceedings from the Sixth International
Date: 1999

Type of Experiement: Case Study



This study investigates whether there is a relationship between Object Oriented metrics and the likelihood of OO faults. The goal is that, if this relationship exists, one can focus the testing on the areas of the code that exhibit these traits. An empirical study was conducted over three industrial real-time systems that contain a number of natural faults that have been reported for the past 3 years. The faults are then categorized into three categories: OO faults, Object Management faults, and traditional faults. The OO metrics suite proposed by Chidamber and Kemerer is validated during these empirical tests and a few additional metrics are proposed to serve as an indicator of how strong an OO program is, so a team can better adopt OO techniques and focus their testing efforts better. The RFC metric was found to be a good indicator of OO faults and WMC can be a good indicator of faulty classes. The new metrics were found to help decide better which classes need to be tested using OO testing techniques. By using WMC as the only fault prone class indicator 36% of faulty classes could be missed and 50% of OO fault prone classes, by using CBM and AMC by themself, 35% and 23% of faulty classes would be missed and 34% and 26% of OO fault classes might be missed. However by combining these 90% of total faults can be discovered and 91% of OO faults can be discovered.