Predicting the Amount of Files Required to Fix a Bug

This paper proposes a classifier that can predict the amount of files required to fix a bug. A newly incoming bug can be classified into one of the three classes (categories): Small; Medium; or Large depending on the amount of files required to fix that bug. For this purpose; 5800 bug reports are studied from three open source projects. The projects are: AspectJ; Tomcat; and SWT. Then; feature sets are extracted for each project separately. The feature sets represent the occurrences of keywords in the summary and description parts of the bug reports. Due to the high dimensionality of the feature vectors; we propose to apply the well-known method; principle component analysis (PCA). The resulting feature vectors are then fed to a number of popular machine learning algorithms. For an enhanced performance; we experiment with multiclass support vector machine quadratic MSVM 2 . It provides improvements of classification accuracy ranging from 2.3% to 22.3% compared to other classifiers.