A Software Defect Prediction Approach Based on Hybrid Feature Dimensionality Reduction
Algorithm 1
HFDRA algorithm.
Input:
Original dataset, FF-Correlation Measurement method: MIC, FC-Correlation measurement method: IG, Dimensions after dimensionality reduction: w, Number of clusters: k;
Output:
Data after dimensionality reduction: R
/ Feature selection stage /
1. Fori = 1 to ndo
2. Forj = 1 to ndo
3. Use the MIC to calculate the correlation and save it in the matrix Fij.
4. end for
5. end for
6. Forh = 1 to ndo
7. Use IG to calculate the and save it to the vector Sh.
8. end for
9. According to the information of the matrix F and the vector C, the K-Medoids algorithm is used to divide the feature set into k clusters.
/ Feature dimensionality reduction stage /
10. Initialize the matrix R
11. Forq = 1 to kdo
12. Use KPCA to reduce the dimensionality of the data subset of cluster Cq to , and add the reduced dimensionality result to R.