Fig. 6

Decision curve analysis (DCA) of the two schemes of reduction. At probability thresholds of 0.0 to 0.9, the RF model based on the LASSO algorithm for TRG classification provided more net benefit than that utilizing PCA (a). Except at the probability threshold of 0.4, with comparable benefits from both models (AUCs of 0.943 and 0.930 respectively). Meanwhile, at threshold probabilities of 0.1 to 0.85, the KNN model based on the LASSO algorithm had increased net benefit than the PCA scheme (b)