An Improved Method for Cross-Project Defect Prediction by Simplifying Training Data

Table 8

Pairwise comparisons between a given combination and each of the 15 combinations in terms of Cliff’s delta () effect size.

	Cosine similarity					Euclidean distance					Manhattan distance
	Linear	Logistic	Square root	Logarithmic	Inverse cotangent	Linear	Logistic	Square root	Logarithmic	Inverse cotangent	Linear	Logistic	Square root	Logarithmic	Inverse cotangent
Cosine + Linear	-	0.018	0.084	0.000	0.116	−0.049	−0.036	−0.004	−0.013	−0.009	0.138	0.049	0.164	0.178	0.169
Euclidean + Linear	0.049	0.102	0.111	0.062	0.164	-	0.036	0.040	0.058	0.089	0.209	0.102	0.249	0.276	0.244
Manhattan + Logistic	−0.049	−0.022	0.022	−0.013	0.111	−0.102	−0.076	−0.080	−0.049	−0.031	0.053	-	0.124	0.151	0.147