Dl. Lanning et Tm. Khoshgoftaar, AN EMPIRICAL-MODEL OF ENHANCEMENT-INDUCED DEFECT ACTIVITY IN SOFTWARE, IEEE transactions on reliability, 44(4), 1995, pp. 672-676
This study exploits the relationship between functional enhancement (F
E) activity and defect distribution to produce a model for predicting
FE induced defect activity, We achieve this in 2 steps: 1) Apply canon
ical correlation analysis to model the relationship between a set of F
E activity indicators and a set of defect activity indicators. This an
alysis isolates 1 dimension of this relationship having strong correla
tion. 2) Model the relationship between the latent variables at this d
imension as a simple linear regression. This model demonstrates predic
tive quality sufficient for application as a software engineering tool
. The predictive model considers FE activity as the sole source of var
iation in defect activity. Other sources of variation, such as differe
nces: in the product to be enhanced, in programmer skill level, in pro
grammer product understanding, and in the software development process
, are not modeled, but remained constant throughout the development ef
fort that yielded the modeled data. Models developed with this techniq
ue are intended for predicting defect activity in the program modules
that result from the next iteration of the same development process, i
n production Of the next release of the modeled product, with the same
key people implementing the software changes that introduce FE. Even
in this application, software engineers should understand & control th
e impacts of the unmodeled sources of variation, The modeling techniqu
e scales to larger development efforts involving several hey people by
either developing unique models for each area of responsibility, or a
dding independent variables that account for variation introduced by d
iffering skill & understanding levels.