Unsupervised software defect prediction using median absolute deviation threshold based spectral classifier on signed Laplacian matrix

Journal of Big Data

Table 4 Clusters membership and performance

Dataset name	Zero threshold based spectral clustering			MAD threshold based spectral clustering
	Cluster memberships (%)		Compactness (DBI)	Cluster memberships (%)		Compactness (DBI)
	\(D_c\)	\(C_c\)	Compactness (DBI)	\(D_c\)	\(C_c\)	Compactness (DBI)
CM1	44.34	55.66	1.3	28.75	71.25	1.1
KC3	44.85	55.15	1.4	21.50	78.50	1.2
MC1	32.80	67.20	2.0	24.60	75.40	1.8
MC2	48.00	52.00	1.9	21.60	78.40	1.1
MW1	45.45	54.55	1.4	30.83	69.17	1.0
PC1	48.23	51.77	1.9	31.49	68.51	1.3
PC2	45.23	54.77	2.2	19.06	80.94	1.5
PC3	48.47	51.53	2.4	30.16	69.84	1.4
PC4	46.23	53.77	1.8	29.91	70.09	1.2
PC5	47.63	52.37	2.1	31.09	68.91	1.9
Average	45.12	54.88	1.8	26.90	73.10	1.4

\(D_c\) is the predicted defective cluster; \(C_c\) is the predicted clean cluster
The italicized values indicate the better performance