Table 2.

AUC values with 95% confidence interval from different ML methods by size of training sample for the systematic review PPU

Training size	Logistic regression	Support vector machine	Random forest	LightGBM	BERT	GPT-2
135	75 [70, 81]	63 [56, 71]	71 [63, 78]	67 [60, 73]	70 [64, 76]	52 [46, 59]
270	72 [67, 78]	60 [51, 68]	69 [62, 76]	73 [67, 80]	79 [74, 84]	54 [48, 60]
405	72 [66, 78]	57 [48, 66]	68 [58, 77]	68 [58, 78]	77 [71, 83]	64 [57, 70]
540	68 [59, 76]	61 [51, 71]	75 [67, 84]	74 [65, 83]	76 [69, 83]	73 [66, 80]
674	69 [61, 76]	60 [50, 70]	77 [68, 86]	80 [72, 87]	91 [88, 94]	76 [69, 83]
808	59 [47, 71]	59 [47, 71]	81 [73, 89]	76 [67, 84]	87 [82, 92]	80 [73, 88]
942	65 [52, 78]	62 [49, 75]	78 [65, 91]	81 [71, 90]	88 [83, 94]	84 [76, 92]
1076	72 [59, 86]	68 [56, 80]	83 [70, 96]	80 [68, 93]	92 [87, 97]	84 [75, 94]
1210	70 [43, 97]	66 [43, 89]	76 [52, 100]	80 [56, 100]	89 [75, 100]	85 [65, 100]

Values are given as a percentage