文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Validating Object-Oriented Design Metrics on a Commercial Java Application
侵权投诉

Validating Object-Oriented Design Metrics on a Commercial Java Application

de l'information

Empirical validation involves demonstrating an association between the metric under study andother measures of important external attributes (ISO/IEC-14598-1, 1996). An external attribute isconcerned with how the product relates to its environment (Fenton, 1991). Examples of externalattributes are testability, reliability and maintainability. Practitioners, whether they are developers,managers, or quality assurance personnel, are really concerned with the external attributes.However, they cannot measure many of the external attributes directly until quite late in aproject’s or even a product’s life cycle. Therefore, they can use product metrics as leadingindicators of the external attributes that are important to them. By having good leading indicators,it is possible to predict the external attributes and take early action if the predictions do not fit aproject’s objectives. For instance, if we know that a certain coupling metric is a good leadingindicator of maintainability as measured in terms of the effort to make a corrective change, thenwe can minimize coupling during design because we know that in doing so we are also increasingmaintainability.

In this paper we focus on empirically validating a set of object-oriented design metrics developedby Chidamber and Kemerer (Chidamber and Kemerer, 1994) and Briand et al. (Briand et al.,1997). The study was performed with data collected from a commercial Java application thatimplements an XML editor. The external attribute that we measure for our study is reliability.Reliability can be measured in different ways. We measure it in terms of the incidence of a post-release fault in a class. This is termed the fault-proneness of a class.

While previous empirical validation studies of these metrics have been performed (Briand et al.,1997; Briand et al., 1998a; Briand et al., 2000; El-Emam et al., 1999, 2000b; El-Emam et al.,2001b), our current work is intended to replicate these studies in order to provide accumulatingevidence of the metrics’ validity. Furthermore, we make a number of additional contributions tothe above studies:

We present a detailed cognitive theory to justify the above metrics. Based on this theory westate a number of precise hypotheses relating the design metrics to fault-proneness.

Evidence supporting the hypotheses suggests possible reasons why certain metrics areleading indicators of fault-proneness. To our knowledge, this is the first attempt at postulatinga detailed cognitive theory for object-oriented metrics.

An object-oriented metrics cost-benefit model is formulated. This model can be used toevaluate the cost savings from using the design metrics to predict which classes are most

fault-prone, and target these classes for inspection. The model is then applied to illustrate itsutility.

Our results indicate that many of the design metrics are indeed associated with fault-proneness.The directions of the associations conform to what we predicted from the cognitive theory. This isencouraging as now we have evidence for some cognitive mechanisms to explain the impact ofobject-oriented metrics. We also built a fault-proneness prediction model using a subset of thevalidated metrics. The fault-proneness prediction model was found to have high accuracy, andthe post-release cost savings from the application of the prediction model were estimated to be42%. This means that had the prediction model been used to identify the classes to inspect (i.e.,those with the highest predicted fault-proneness), then 42% of the post-release costs (i.e., costsassociated with dealing with faults discovered by customers) would have been saved. For manyorganizations such savings are nontrivial, and can free up resources to add features to theirproducts.

In the next section we present a theory that provides mechanisms explaining why the object-oriented metrics we study would be associated with fault-proneness, and give an overview of themetrics we study. Section 3 presents our research method in detail, and our results arepresented and discussed in Section 4. We conclude the paper in Section 5 with a summary anddirections for future work.

上一页第4页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email