Table 7.

Performance comparison of the proposed models with respect to the literature.

Papers	Overall dataset	Features	Methods	Sampling	Best
	(recurrence/no recurrence)			strategy	performances (%)
Mohebian et al. (8)	Private database	Patient	Particle Swarm	Hold-out	AUC: 90.0%
	(112/467)	and histopathological	Optimization		Acc: 90.0%
		characteristics,	and Bagged		Sens: 81.0%
	Follow-up 5 year	and therapy	Decision Tree		Spec: 98.0%
Beheshti et al. (45)	Wisconsin Prognostic BC	Textural	Genetic	Hold-out	AUC: 63.0%
	(UCI Repository)	and histopathological	Programming		Acc: 80.3%
	(47/151)	characteristics	approach		Sens: 52.3%
	Follow-up 5 year				Spec: 83.4%
Chaurasia and Pal (9)	University Medical Centre,	Patient	Simple Logistic	10-fold	AUC: -%
	Ljubljana, Yugoslavia	and histopathological		cross validation	Acc: 74.4%
	(85/201)	characteristics,			Sens: 31.8%
	Follow-up 5 year	and therapy			Spec: 92.5%
Kim et al. (11)	Private database	Histopathological	SVM	Hold-out	AUC: 85.0
	(195/484)	characteristics			Acc: 84.6%
		and therapy			Sens: 89.0%
	Follow-up 5 year				Spec: 73.0%
Proposed models	Private Database	Patient	SVM-RFE feature	10-fold	AUC: 89.1%
	(54/159)	and histopathological	selection and	cross-validation	Acc: 78.8%
		characteristics,	RF classifier		Sens: 86.6%
	Follow-up 5 years	and therapy			Spec: 76.4%
Proposed models	Private database	Patient	SVM-RFE feature	Hold-out	AUC: 87.8%
	(54/159)	and histopathological	selection and		Acc: 77.5%
		characteristics,	RF classifier		Sens: 92.3%
	Follow-up 5 years	and therapy			Spec: 70.4%