Evaluating the Impact of Data Transformation Techniques on the Performance and Interpretability of Software Defect Prediction Models

Table 4

Summary of the impact of the transformation methods studied on the performance and interpretability of different SDP models.

Methods	Models	Performance increase	Importance change
Log	RF	Recall	None
	LR	Accuracy, Recall, F1, AUC, MCC	Top-1, 3, 5
	DT	AUC	None
	NB	Recall, F1, AUC, MCC	Top-1, 3, 5
	KNN	Accuracy, Precision, Recall, F1, AUC, MCC	Top-1, 3, 5
	MLP	Recall, F1	Top-1, 3, 5
Minmax	RF	None	None
	LR	Accuracy	Top-1, 3, 5
	DT	AUC	None
	NB	Accuracy, Recall, F1, AUC, MCC	None
	KNN	Precision	Top-1, 3, 5
	MLP	Accuracy, Precision, F1, AUC, MCC	Top-1, 3, 5
Z-score	RF	None	None
	LR	Accuracy, Precision, F1, AUC, MCC	Top-1, 3, 5
	DT	AUC	None
	NB	Accuracy, Recall, F1, AUC, MCC	None
	KNN	Precision	Top-1, 3, 5
	MLP	Accuracy, Precision, Recall, F1, AUC, MCC	Top-1, 3, 5

Note. Only significantly improved indicators are displayed. None indicates that there are no significantly improved indicators or no feature importance changes.