Skip to main content

Table 2 Performances of RF, BRT, LASSO and UFMLR in the 500 simulated datasets

From: Detection of independent associations in a large epidemiologic dataset: a comparison of random forests, boosted regression trees, conventional and penalized logistic regression for identifying independent factors associated with H1N1pdm influenza infections

  n RF BRT LASSO-max LASSO-se UFMLR05 UFMLR05 backward UFMLR20 UFMLR20 backward
Type I error 5%          
True Positive Rates (TPR) 8 85% (55% - 100%) 80% (51% - 100%) 78% (52% - 100%) 71% (41% - 100%) 28% (3% - 54%) 45% (20% - 70%) 26% (0% - 65%) 49% (15% - 84%)
Covariates with pairwise interaction 4 86% (49% - 100%) 80% (41% - 100%) 77% (40% - 100%) 69% (26% - 100%) 24% (0% - 63%) 41% (0% - 83%) 24% (0% - 72%) 46% (0% - 96%)
Covariates without pairwise interaction 4 84% (46% - 100%) 79% (41% - 100%) 79% (41% - 100%) 73% (32% - 100%) 32% (0% - 74%) 50% (6% - 93%) 28% (0% - 76%) 53% (5% - 100%)
Continuous covariates 4 90% (55% - 100%) 82% (47% - 100%) 82% (49% - 100%) 77% (41% - 100%) 35% (0% - 74%) 49% (15% - 84%) 29% (0% - 76%) 55% (14% - 95%)
Binary covariates 4 80% (34% - 100%) 78% (35% - 100%) 74% (34% - 100%) 64% (20% - 100%) 22% (0% - 57%) 41% (8% - 74%) 23% (0% - 69%) 44% (0% - 90%)
False Positive Rates (FPR) 292 4% (1% - 6%) 4% (2% - 6%) 9% (0% - 17%) 4% (0% - 9%) 2% (1% - 4%) 3% (1% - 5%) 4% (0% - 12%) 9% (0% - 18%)
Covariates correlated with associated covariates 12 46% (7% - 85%) 33% (3% - 63%) 23% (0% - 48%) 18% (0% - 41%) 4% (0% - 15%) 7% (0% - 22%) 9% (0% - 39%) 19% (0% - 60%)
Covariates uncorrelated with associated covariates 280 2% (0% - 4%) 3% (1% - 5%) 8% (0% - 17%) 3% (0% - 8%) 2% (1% - 4%) 3% (1% - 4%) 4% (0% - 11%) 8% (0% - 16%)
Continuous covariates 146 3% (0% - 6%) 3% (1% - 6%) 9% (0% - 18%) 4% (0% - 9%) 2% (0% - 5%) 3% (0% - 6%) 5% (0% - 13%) 9% (0% - 18%)
Binary covariates 146 4% (0% - 8%) 5% (2% - 9%) 9% (0% - 18%) 3% (0% - 9%) 2% (0% - 5%) 3% (0% - 5%) 4% (0% - 12%) 9% (0% - 18%)
  1. Performances are shown as: Mean (95% confidence interval).