Research on imbalance machine learning methods for MR $$T_1$$ WI soft tissue sarcoma data

Liu, Xuanxuan; Guo, Li; Wang, Hexiang; Guo, Jia; Yang, Shifeng; Duan, Lisha

doi:10.1186/s12880-022-00876-5

BMC Medical Imaging

Table 6 Performance of the SRS dataset splitting method on 17 models in the testing set

From: Research on imbalance machine learning methods for MR$T_1$WI soft tissue sarcoma data

N	FS	ST	CM	AUC ± ${\sigma }$	Acc(%) ± ${\sigma }$	Sens(%) ± ${\sigma }$	Spec(%) ± ${\sigma }$	G-mean ± ${\sigma }$
1	RFE	ROSE	ERT	0.9308 ± 0.0445	95.49 ± 0.0245	98.69 ± 0.0139	87.47 ± 0.0915	0.9278 ± 0.0478
2	RFE	SMOTE	ERT	0.9438 ± 0.0382	95.69 ± 0.0203	96.66 ± 0.0229	92.10 ± 0.0713	0.9429 ± 0.0391
3	RFE	STT	ERT	0.9438 ± 0.0382	95.69 ± 0.0203	96.66 ± 0.0229	92.10 ± 0.0713	0.9429 ± 0.0391
4	RFE	ADASYN	ERT	0.9419 ± 0.0430	94.90 ± 0.0280	96.11 ± 0.0200	92.28 ± 0.0821	0.9409 ± 0.0443
5	RFE	ROSE	RF	0.9358 ± 0.0410	94.71 ± 0.0321	96.30 ± 0.0363	90.86 ± 0.0754	0.9345 ± 0.0425
6	RFE	SMOTE	RF	0.9087 ± 0.0547	92.94 ± 0.0211	94.24 ± 0.0326	87.49 ± 0.1104	0.9059 ± 0.0600
7	RFE	STT	RF	0.9197 ± 0.0439	93.14 ± 0.0349	94.31 ± 0.0420	89.63 ± 0.0759	0.9185 ± 0.0447
8	RFE	ADASYN	RF	0.9220 ± 0.0429	92.55 ± 0.0289	92.62 ± 0.0322	91.78 ± 0.0865	0.9208 ± 0.0437
9	RFE	ROSE	BRF	0.9356 ± 0.0396	94.90 ± 0.0295	96.86 ± 0.0324	90.27 ± 0.0749	0.9342 ± 0.0412
10	RFE	SMOTE	BRF	0.9111 ± 0.0562	93.14 ± 0.0212	94.23 ± 0.0229	88.00 ± 0.1095	0.9088 ± 0.0614
11	RFE	STT	BRF	0.9350 ± 0.0284	93.53 ± 0.0186	93.89 ± 0.0315	93.10 ± 0.0695	0.9339 ± 0.0291
12	RFE	ADASYN	BRF	0.9388 ± 0.0404	93.73 ± 0.0304	93.99 ± 0.0346	93.78 ± 0.0791	0.9378 ± 0.0415
13	RFE	ROSE	SVM	0.8191 ± 0.0448	87.84 ± 0.0180	94.53 ± 0.0324	69.29 ± 0.1085	0.8062 ± 0.0559
14	RFE	SMOTE	SVM	0.8276 ± 0.0545	86.08 ± 0.0339	88.94 ± 0.0445	76.59 ± 0.1120	0.8227 ± 0.0586
15	RFE	STT	SVM	0.8276 ± 0.0545	86.08 ± 0.0339	88.94 ± 0.0445	76.59 ± 0.1120	0.8227 ± 0.0586
16	RFE	ADASYN	SVM	0.8699 ± 0.0573	89.22 ± 0.0349	90.99 ± 0.0374	83.00 ± 0.1204	0.8664 ± 0.0604
17	RFE	GDO	SVM	0.8143 ± 0.0598	87.06 ± 0.0230	92.34 ± 0.0373	70.52 ± 0.1395	0.8020 ± 0.0752

Best results are highlighted in bold style
N number; FS feature selection; ST sampling technique; CM classification method; AUC area under the curve; Sens sensitivity; Spec: specificity; ROSE random oversampling examples; SMOTE synthetic minority oversampling technique; STT SMOTETomek; ADASYN adaptive synthetic RFEsampling; RFE recursive feature elimination; ERTextremely randomized trees; RFrandom forest; BRF balanced random forest; SVM support vector machine

Back to article page

ISSN: 1471-2342

Contact us

General enquiries: journalsubmissions@springernature.com