The practical clinical role of machine learning models with different algorithms in predicting prostate cancer local recurrence after radical prostatectomy

Table 5 Predictive performance of different models in training and testing sets

Models		AUC (95% CI)	Accuracy	Sensitivity	Specificity	PPV	NPV
LR-LASSO	Training	0.873 (0.807–0.938)	0.789	0.763	0.800	0.630	0.883
	Testing	0.858 (0.746–0.971)	0.849	0.750	0.892	0.750	0.892
SVM	Training	0.875 (0.808–0.942)	0.813	0.737	0.847	0.683	0.878
	Testing	0.829 (0.708–0.951)	0.755	0.500	0.865	0.615	0.800
LDA	Training	0.885 (0.824–0.946)	0.821	0.737	0.859	0.700	0.880
	Testing	0.845 (0.742–0.948)	0.717	0.500	0.811	0.533	0.790
PI-RR	Training	0.849 (0.770–0.929)	0.829	0.710	0.882	0.730	0.872
	Testing	0.833 (0.708–0.958)	0.811	0.625	0.892	0.714	0.846
Combined	Training	0.907 (0.853–0.961)	0.805	0.895	0.765	0.630	0.942
	Testing	0.924 (0.851–0.997)	0.868	0.875	0.865	0.737	0.941

AUC = area under the curve; CI = confidence interval; PPV = positive predictive value; NPV = negative predictive value;
LR-LASSO = logistic regression-least absolute shrinkage and selection operator; SVM = support vector machine; LDA = linear discriminant analysis; PI-RR = Prostate Imaging for Recurrence Reporting

ISSN: 1470-7330