Improvement of causal analysis using multivariate statistical process control

Abstract

Statistical process control (SPC) is a conventional means of monitoring software processes and detecting related problems, where the causes of detected problems can be identified using causal analysis. Determining the actual causes of reported problems requires significant effort due to the large number of possible causes. This study presents an approach to detect problems and identify the causes of problems using multivariate SPC. This proposed method can be applied to monitor multiple measures of software process simultaneously. The measures which are detected as the major impacts to the outof-control signals can be used to identify the causes where the partial least squares (PLS) and statistical hypothesis testing are utilized to validate the identified causes of problems in this study. The main advantage of the proposed approach is that the correlated indices can be monitored simultaneously to facilitate the causal analysis of a software process.

Establishing a Defect Management Process Model for Software Quality Improvement

“…It will be more effective to perform the FMEA and FTA at the start of the project this will reduce the time and cost spend to fix the defects in the later stages. In CMM at level5 defect management is considered as a key process area to plan defect prevention activities [9].…”

Integrating Action-Based Defect Prediction to Provide Recommendations for Defect Action Correction

“…The challenge to analyze the selected items is the possibility that a broad range of causes may exist, requiring signi¯cant e®ort to identify the actual causes. The Multilevel Software Cause Identi¯cation (MSPC) approach groups measures into multiple levels to detect problems of software process and identify the causes of problems, in which the partial least squares (PLS) regression method and statistical hypothesis testing are utilized to validate the identi¯ed causes [21]. The MSPC is applied to identify the attributes that are likely to cause the prede¯ned problems.…”

“…Two numeric patterns, p 1 ¼ fid 1 : ½v 11 ; v 12 g and p 2 ¼ fid 2 : ½v 21 ; [3;5] are partially jointed. Based on the de¯nitions, the patterns of mined rules can be analyzed to identify the ACR.…”

“…7 shows that the intervals I 12 ¼ ½v 11 ; v 12 and I 22 ¼ ½v 21 ; v 22 are overlapping. The value v 2 (the value of b 2 ) can fall in the overlap region I C ¼ ½v 21 ; v 12 or the high region I H ¼ ½v 11 ; v 21 (at the left of the I C ), the region with high characteristics. The region I L ¼ ½v 12 ; v 22 (at the right of the I C ) is the region with low characteristics.…”

“…To appease the high possibility of causing problems, v 2 of action a k is suggested to change to the region I L (the ACRI of b 2 ). Case 3: When I 12 and I 22 are partially disjointed (assume that v 12 < v 21 ), the intervals I 12 ¼ ½v 11 ; v 12 and I 22 ¼ ½v 21 ; v 22 have no common region. Figure 8 shows that the high characteristic of the action a k can be degraded by changing v 1 to the interval I L ¼ I 22 .…”