Fig. 4
From: Predicting self-perceived general health status using machine learning: an external exposome study
Effect of increasing the number of variables within RF models on the AUC value. Legend: Left panel a shows results from models based on the 2012 dataset and right panel b of models based on thee2016 dataset. The AUC values represent the average AUC based on fivefold cross validation and its 95% CI in RF models with an increasing number of variables inside the RF models. Default settings of the ranger package were used in the assembly of each RF model. Variables are added in an order that corresponds to the variable importance ranking of the complete model. The dashed line represents the average AUC value of the RF forest containing all variables (2012: n = 81; 2016: n = 91)