A Systematic Review of the Empirical Validation of Object-Oriented Metrics Towards Fault-Proneness Prediction

Abstract

Object-oriented (OO) approaches of software development promised better maintainable and reusable systems, but the complexity resulting from its features usually introduce some faults that are di±cult to detect or anticipate during software change process. Thus, the earlier they are detected, found and¯xed, the lesser the maintenance costs. Several OO metrics have been proposed for assessing the quality of OO design and code and several empirical studies have been undertaken to validate the impact of OO metrics on fault proneness (FP). The question now is which metrics are useful in measuring the FP of OO classes? Consequently, we investigate the existing empirical validation of CK þ SLOC metrics based on their state of signi¯cance, validation and usefulness. We used systematic literature review (SLR) methodology over a number of relevant article sources, and our results show the existence of 29 relevant empirical studies. Further analysis indicates that coupling, complexity and size measures have strong impact on FP of OO classes. Based on the results, we therefore conclude that these metrics can be used as good predictors for building quality fault models when that could assist in focusing resources on high risk components that are liable to cause system failures, when only CK þ SLOC metrics are used.

Analysis of Metric-Based Object-Oriented Code Refactoring Opportunities Identification Approaches

“…In particu lar, OO product metrics captures different structural features of OO software systems such as class complexity, coupling and cohesion [7] [15]. Today, several OO metrics exist and emp irically validated in the assessment of OO design and codes quality [32] [33]. In the 16 studies analyzed in this paper, we have seen the impact of software met rics in the assessment and improvement of software quality.…”

A Framework for Effective Object-Oriented Software Change Impact Analysis

“…In this case, the quality of the systems can be assessed via pure prediction using software metrics such as code metrics, past change and fault histories as well as suitable fault prediction model. Several empirical studies in the literature have confirmed the relationship between product and process metrics and fault-proneness [10,11,12]. To carry out the prediction, all the measures extracted from either the previous or current version of the software stored in the database will be used to predict whether a component affected by a change will be faulty or not.…”

“…The motivation is that software quality is known to play a crucial role in the success and failure of any software organization. However, in large software systems, providing high quality in development has been deemed complex and a laborious activity [12]. In this case, it is important that the available resources are focused on the most critical parts of the system to ensure customers' satisfaction.…”

“…However, these CIA approaches provide little or no information on how to represent OO program for effective maintenance. Furthermore, OO classes are not faults or failures free [1,2,3,4,10,11,12,13,14]. Obviously, the two activities are carried out separately and there are no known approaches of CIA that incorporates both change impact and fault prediction.…”

“…Thus, the intuition is that if a fault-prone class is changed without fixing the existing faults, it may increase the efforts and costs of the maintenance or could lead to software failure. In the realm of project management, time, cost and scope constitutes the three "stalagmites" where quality is the goal [1] [12]. As faults during development are inevitable, the earlier they are found and fixed, the lesser it costs and the higher the quality of the products delivered [1,2,3].…”