Benefits of Principal Components Analysis with Classification Trees of Fault-Prone Software Modules

摘要

Software reliability has become a major concern in many software-intensive industries, because faults in source code can cause operational failures with serious consequences. This paper confirms prior work showing that classification trees can be useful to identify fault-prone modules based on the pattern of software metrics and furthermore, shows that principal components analysis can be a useful supporting technique. We conducted a case study of a very large telecommunications system. Tree models built by the Classification And Regression Trees (CART) algorithm illustrated the potential benefits of preprocessing data with principal components analysis.