Skip to main content

Table 4 Prediction models assessed with Neural Network approach

From: Prediction of steroid resistance and steroid dependence in nephrotic syndrome children

Prediction parameter

Neural Network

Train a

Test b

NS prediction Sensitivity %

75% (75/100)

38% (9/24)

NS prediction Specificity %

80% (35/44)

91% (10/11)

Total number of correct calls %

76% (110/144)

54% (19/35)

Log Loss

0.483

0.682

AUC

0.825

0.561

SR prediction Sensitivity %

68% (34/50)

33% (1/3)

SR prediction Specificity %

98% (49/50)

100% (21/21)

Total number of correct calls %

83% (83/100)

92% (22/24)

Log Loss

0.391

0.72

AUC

0.932

0.778

SD prediction Sensitivity %

62% (16/26)

80% (4/5)

SD prediction Specificity %

91% (32/35)

80% (4/5)

Total number of correct calls %

79% (48/61)

80% (8/10)

Log Loss

0.704

1.208

AUC

0.866

0.72

SR prediction Sensitivity %

82% (28/34)

63% (12/19)

SD prediction Sensitivity %

46% (13/28)

67% (2/3)

PSS prediction Sensitivity %

63% (24/38)

50% (1/2)

Total number of correct calls %

65% (65/100)

63% (15/24)

Log Loss

0.756

0.761

  1. NS nephrotic syndrome, SR steroid resistant, SS steroid sensitive, SD steroid dependent, PSS primarily steroid sensitive, AUC area under the curve
  2. Each model was run using aTrain set, comprising 80% of data, and bTest set, comprising the rest 20% of data