The authors of this paper claim to describe a practical method that employs machine learning techniques to identify potentially unreliable software modules during their development, by using data obtained from already-finished modules. The presented results appear to show that their decision tree method (or possibly some variation of it) generated predictions similar to those generated using another method. No attempt was made either to explain the differences in classification between the two models or to quantify the cost of the new method. The proposed method requires training—this was performed once using around 90 percent of the software modules making up a large medical software system—though no results are provided to show how the method performed as an evolutionary procedure (that is, how the accuracy of prediction varied as the number of modules available increased).

The paper is very poorly constructed with a number of introductory sections, such as one on software reliability and software metrics, that offer little or nothing to readers at any level of expertise. These sections contain a number of misleading and garbled statements; for example, “It is difficult to estimate the time and effort it will take to remove all defects,” and “It has been shown that the pattern of the faults insertion phenomenon is related to measurable attributes of the software, especially with the software metrics.”

This paper cannot be recommended as useful to any group of readers. It certainly does not add anything to the field of medical software development, nor does it add anything to the general field of software engineering. I recommend that the journal publishing this article overhaul its acceptance procedures to prevent papers of such a poor quality from being published in the future.