Using Attribute Focusing to Diagnose and Correct the Software Production Process

Software production laboratories classify and track defect data. As defects represent inadequacies in the definition or the execution of the production process, these data must be analyzed to improve the process. Traditionally, the identification and correction of such process inadequacies (called process feedback) has largely been a difficult, time-consuming, manual exercise. We seek to automate process feedback. An obvious approach is to identify patterns in defect data that are characteristic of common process inadequacies, and then take appropriate action to remedy the process. The problem with this approach is that it assumes knowledge of process inadequacies. Clearly, given such knowledge one would simply start with the correct process and prevent defects from occurring in the first place. This suggests that major problems with the quality of a software product occur because of process inadequacies that are not known till it is much too late to correct the process. To address process inadequacies, we have devised Attribute Focusing (AF), an automatic technique to discover unusual trends in classified defect data. AF does not match a knowledge base of patterns against the data. Instead, it computes patterns of attribute-values that are deemed to be unusual based on a degree of magnitude or association. Once discovered, the unusual patterns must be explained by the project team. These explanations lead to the identification of process inadequacies which can then be corrected. In other words, AF incorporates a model of interestingness to select patterns that should lead the project team to learn about their process.