Table 1.

Predictive accuracy of the random forests classifier for identifying pain cries vs. hungry vs. fussy cries, assessed using the out-of-sample accuracy

Calculated diagnostic accuracy parameters
Sample size	691
Prevalence	0.51
Sensitivity	0.91
Specificity	0.68
PPV	0.75
NPV	0.87
LR + result	2.81
LR − result	0.14

The primary algorithm was trained on these three cry states that were not developmentally dependent, to assess whether pain ratings differed in babies with colic and without colic. Roughly 51% of cries were painful, but the ChatterBaby algorithm performed significantly above chance and correctly flagged 91% of pain cries