Skip to main content

Table 4 Performance of the probability calibration-based algorithms

From: Probability calibration-based prediction of recurrence rate in patients with diffuse large B-cell lymphoma

  ECE MCE BS P_value
NB 15.711 (13.557–17.914) 34.350 (29.275–39.800) 0.212 (0.199–0.228) < 0.001(< 0.001- < 0.001)
NB-Platt 9.008 (7.919–10.647) 21.550 (17.475–25.800) 0.189 (0.181–0.197) 0.179 (0.055–0.389)
NB-IsoReg 9.820 (7.740–12.190) 40.000 (23.475–57.100) 0.208 (0.195–0.227) < 0.001(< 0.001–0.057)
NB-RPR 8.743 (7.397–10.307) 21.600 (17.575–25.700) 0.189 (0.182–0.197) 0.191 (0.051–0.431)
LR 8.517 (7.244–10.093) 20.100 (16.675–25.025) 0.188 (0.180–0.196) 0.152 (0.022–0.403)
LR-Platt 8.981 (7.478–10.485) 20.900 (17.300–25.325) 0.189 (0.182–0.196) 0.215 (0.065–0.437)
LR-IsoReg 9.140 (6.970–11.810) 31.550 (20.000–50.175) 0.204 (0.193–0.220) 0.008(< 0.001–0.348)
LR-RPR 8.744 (7.308–10.143) 20.300 (16.700–24.425) 0.187 (0.181–0.194) 0.255 (0.092–0.507)
RF 12.740 (10.910–14.336) 27.200 (23.375–31.925) 0.201 (0.190–0.211) < 0.001(< 0.001- < 0.001)
RF-Platt 8.998 (7.518–10.447) 21.100 (17.500–26.700) 0.192 (0.184–0.200) 0.156 (0.030–0.435)
RF-IsoReg 9.292 (7.332–11.353) 27.850 (20.000–40.000) 0.201 (0.191–0.215) < 0.001(< 0.001–0.131)
RF-RPR 8.949 (7.387–10.524) 20.900 (17.400–26.025) 0.189 (0.182–0.196) 0.194 (0.061–0.458)
SVM 9.872 (8.317–11.777) 23.800 (19.000–28.925) 0.194 (0.185–0.204) 0.016(< 0.001–0.117)
SVM-Platt 9.077 (7.702–10.895) 21.750 (17.600–27.300) 0.192 (0.184–0.201) 0.169 (0.029–0.412)
SVM-IsoReg 9.501 (7.332–12.453) 30.350 (20.000–42.200) 0.205 (0.194–0.221) 0.003(< 0.001–0.249)
SVM-RPR 8.796 (7.362–10.439) 21.000 (16.775–26.550) 0.190 (0.183–0.199) 0.211 (0.064–0.471)
FFNN 8.238 (6.805–9.611) 20.150 (16.600–24.500) 0.184 (0.177–0.192) 0.244 (0.075–0.518)
FFNN-Platt 8.991 (7.721–10.642) 20.950 (16.875–26.100) 0.186 (0.179–0.194) 0.192 (0.056–0.425)
FFNN-IsoReg 10.866 (8.603–13.347) 40.550 (27.800–57.025) 0.211 (0.196–0.230) < 0.001(< 0.001–0.003)
FFNN-RPR 8.703 (7.393–10.361) 21.400 (17.700–26.025) 0.185 (0.178–0.193) 0.227 (0.073–0.473)
  1. NB naïve Bayes, LR logistic regression, RF random forest, SVM support vector machine, FFNN feedforward neural network, Platt Platt scaling, IsoReg isotonic regression, RPR shape-restricted polynomial regression. “-Platt”, “-IsoReg” and “-RPR” represent performing probability calibration by using corresponding method. In each cell M(P25 - P75): M is the median, P25 is the 25th percentile and P75 is the 75th percentile of 500 evaluations. For each algorithm, the best performance in each column is in bold