Skip to main content

Table 1 Fivefold cross-validation results’ summary of forward predictive power and prediction range comparison between 2D and 3D stochastic models

From: 3D kernel-density stochastic model for more personalized glycaemic control: development and in-silico validation

  Total predictions 1-hourly 2-hourly 3-hourly
58,539 57,840 57,141
2D model % predictions in 25th–75th 55.9 53.4 52.6
% predictions in 5th–95th 91.4 91.0 91.0
3D model % predictions in 25th–75th 52.6 51.3 51.0
% predictions in 5th–95th 90.5 90.2 90.2
3D vs. 2D model % of tighter predictions using 3D model 73.8 72.8 69.9
% reduction in 5th–95th prediction width 24.4 [17.7 29.4] 17.9 [10.9 20.9] 15.5 [10.8 19.2]
% of wider predictions using 3D model 26.2 27.2 30.1
% increase in 5th–95th prediction width 22.0 [7.5 49.1] 16.4 [7.7 32.0] 14.8 [6.8 28.2]
  1. Data given as median [IQR] where appropriate