Research on Cross-Company Defect Prediction Method to Improve Software Security

Algorithm 1

Metric clustering algorithm.

Input: source and target multigranularity metric feature vector and ; number of clusters K;
Output: source and target representative vector and ;
(1)	For each project {Source project S, Target project T}
(2)	Randomly select K metrics as the starting centroid of the project ;
(3)	Repeat the following process until convergence
(4)	For each metric :
(5)	Calculate the Euclidean distance to each starting centroid based on eq.(2);
(6)	Assign the metric to its nearest cluster with minimum distance;
(7)	End for
(8)	For each cluster :
(9)	Calculate the mean value of the cluster and update its centroid;
(10)	End for
(11)	For each cluster :
(12)	Use PCA method to extract the corresponding representative vector ;
(13)	End for
(14)	End for
(15)	Output and ;