Quality metric development framework

Journal of Computer Science, July, 2005 by K. Mustafa, R.A. Khan

Abstract: Several object-oriented metrics have been developed and used in conjunction with the quality models to predict the overall quality of software. However, it may not be enough to propose metrics. The fundamental question may be of their validity, utility and reliability. It may be much significant to be sure that these metrics are really useful and for that their construct validity must be assured. Thereby, good quality metrics must be developed using a foolproof and sound framework / model. A critical review of literature on the attempts in this regard reveals that there is no standard framework or model available for such an important activity. This study presents a framework for the quality metric development called Metric Development Framework ([sup.q]MDF), which is prescriptive in nature. [sup.q]MDF is a general framework but it has been established specially with ideas of object-oriented metrics. [sup.q]MDF has been implemented to develop a good quality design metric, as a validation of proposed framework. Finally, it is defended that adaptation of [sup.q]MDF by metric developers would yield good quality metrics, while ensuring their construct validity, utility, reliability and reduced developmental effort.

Keywords: Object Oriented Metrics, Quality Models, OO Paradigm, OO Structure

INTRODUCTION

Several research works in the object oriented design metrics arena were produced in recent years' [1-9, 14, 17, 31-35]. However, widespread adaptation of object oriented metrics in numerous application domains should only take place if metrics may be shown to be theoretically valid, in the sense that they accurately measure the attributes of software for which they were designed to measure and have also been validated empirically. But there is a general agreement among the experts that metrics have not been even validated theoretically and what to talk of experimental validation. Most of the metrics are accepted by practitioners on 'heavy usages and popularity' and by academic experts on empirical validation. In such a scenario many of the available metrics may not be used properly and have been discarded. Reasons may be that these could not find a place among practitioners or not found to be valid by experts on empirical findings. Now the question is 'Why such metrics which could not become acceptable, were developed?'- leading to all the efforts on development going vain. Therefore, there appears a need for such development to be sound enough and backed by valid procedures to avoid such embarrassing situations. That is, a sound and standard framework or model for metrics development may be quite useful. And that it could lead to the development of good quality metrics.

Development of object oriented metrics: It is evident from the review of literature that few of the researchers have proposed the criteria for developing the desired metrics. We could not explore a standard framework or model for designing object oriented metrics. Therefore, it appeared worthwhile for looking at major developments and discover the direct or indirect use of criteria, methods, guideline etc. Some of the major attempts have been critically reviewed (SATC's Approach [11], Jagdish Bansiya and Carl Devis's Attempt [2], Kitchenham's Approach [12] Abreu's Approach [14] and Victor's Approach [15] and the feeling gets confirmed that there are no known comprehensive and complete models or frameworks or concerned approach that may be used to design quality object oriented metrics. In order to further assure the conclusion, a question regarding the availability of standard framework for designing the desired metrics was posed to various researchers and practitioners. Their responses were analyzed and conclusion was drawn that no such framework is available, and no major attempts on its development has been reported. Despite the apparent diversity of object oriented design framework, an opportunity to work out a desired framework for designing the object oriented metrics is knocking. The need to have a framework that may be used for development of quality metrics motivates the proposal of an appropriate framework. However, a set of common and desired features for development of object oriented metrics can be drawn from the reported attempts on development of metrics. Applying object oriented metric to new paradigms such as object oriented domain has been a dominant feature of academic research up to the present day [16, 17]. The researchers community has developed a considerable number of object oriented metrics. The basic premise behind the development of object oriented metrics is that they can serve as early predictors of classes that contain faults or that are costly to maintain. CK suit is the most referred and most commercially used metrics collection tool available. Dr. Linda Rosenberg [11] at SATC validated the six CK metrics and found that the main features the metrics are incorporating the coverage of all the object oriented concepts. External complexity and internal object structure was found to be the much desired feature for designing the object oriented metrics [11]. Victor Laing [15] with Dr. Linda Rosenberg working for NASA at SATC suggest Orthogonality as one of the important characteristics of the desired metrics to be used for object oriented technology to find the minimal set of metrics. Abreu [18] proposed MOOD set of metrics, which allow the use of the attributes of the object oriented paradigm to be evaluated and reviewed. Abreu strongly suggests that metrics definition and dimension should be justified as it plays an important role in designing the object oriented metrics [14]. System size independence and language independence was also found to be the major contributor for designing the desired metrics [14]. Kitchenham's work in the direction of designing the object oriented metrics reveal that the dimensional consistency and use of correct unit and scale type be the essential feature of the design metrics [12]. It was suggested that the developed metrics must preserve all intuitive notions about the attributes and the way in which the metrics distinguish between entities [13].