Kolmogorov Smirnov and Scatter Matrix Vector for Software Fault Prediction

Many proposed methods for detecting harmful soft
ware are designed based on software quality metrics. Here, bad
smell refers to indicators of potential software code problems that
lead to faults. Moreover, most of the other automated detection
methods struggle with limited accuracy. This work proposes a
novel technique called Kolmogorov Smirnov Refactoring and
Class Balanced Scatter Matrix Vector (KSR-CBSMV) for soft
ware fault prediction. The KSR-CBSMV method is split into
two processes. First, with the JAVA packages provided as input,
JAVA classes and significant software metrics are obtained by
employing Kolmogorov-Smirnov-based Wilcoxon Ranking and
Fisher Statistical Refactoring model. Next, with the extracted
software metrics, software fault prediction is made using Class
Balanced Scatter Matrix-based Support Vector Classifier. Finally,
the results of the JAVA classes identified with defects or not are
predicted promptly and accurately. The experiments show that
our KSR-CBSMV method performs significantly better than two
bug report predictions regarding three performance indicators:
false positive rate, prediction time and software fault prediction
accuracy.
Index Terms—Kolmogorov Smirnov, Wilcoxon Ranking, Sta
tistical Refactoring, Software Fault Prediction, Class Imbalance,
Scatter Matrix and Support Vector Machine