Fig. 5: Comparative BN-logistic and traditional Bayesian Network parameter learning (i.e., Maximum Likelihood Estimation and Bayesian estimation) on simulated data using calibration curve intercept, calibration curve slope, AUC, sensitivity, specificity, and Youden’s index. | npj Digital Medicine

Fig. 5: Comparative BN-logistic and traditional Bayesian Network parameter learning (i.e., Maximum Likelihood Estimation and Bayesian estimation) on simulated data using calibration curve intercept, calibration curve slope, AUC, sensitivity, specificity, and Youden’s index.

From: Interpretable machine learning model for digital lung cancer prescreening in Chinese populations with missing data

Fig. 5

a Calibration curve intercept, which is optimal when close to zero. The red dashed line represents the reference line at zero. b Calibration curve slope, with optimal values close to one, denoted by a red dashed line at one. c represents AUC; d indicates sensitivity; e represents specificity; and f denotes Youden’s index. Higher values of AUC, sensitivity, specificity, and Youden’s index are preferable. The figure displays the results of 1000 statistical simulations. Details on the simulation can be found in Supplementary Note 6. After verifying that each metric approximately follows a normal distribution, the paired t-test was employed to assess whether the BN-logistic model outperformed traditional methods. A p-value of <0.05 indicates superior performance of BN-logistic.

Back to article page