Table 2 Classification accuracy metrics for daily and weekly models (Ndepr = 74,990, Nptsd = 54,197).

From: Forecasting the onset and course of mental illness with Twitter data

Depression MVR \(\mu \) DC \(\mu \) Daily \(\mu (\sigma )\) Weekly \(\mu (\sigma )\)
Recall 0.510 0.614 0.518 (0.000) 0.521 (0.000)
Specificity 0.813 N/A 0.958 (0.000) 0.969 (0.000)
Precision 0.42 0.742 0.852 (0.000) 0.866 (0.000)
NPV 0.858 N/A 0.812 (0.000) 0.841 (0.000)
F1 0.461 0.672 0.644 (0.000) 0.651 (0.000)
PTSD TBA \(\mu \) NHC \(\mu \) Daily \(\mu (\sigma )\) Weekly \(\mu (\sigma )\)
Recall 0.249 0.82 0.683 (0.000) 0.658 (0.000)
Specificity 0.979 N/A 0.988 (0.000) 0.994 (0.000)
Precision 0.429 0.86 0.882 (0.000) 0.934 (0.000)
NPV 0.602 N/A 0.959 (0.000) 0.954 (0.000)
F1 0.315 0.84 0.769 (0.000) 0.772 (0.000)
  1. Accuracy scores from Mitchell et al.28 (MVR), De Choudhury et al.8 (DC), Taubman-Ben-Ari et al.22 (TBA), and Nadeem, Horn, & Coppersmith13 (NHC) are included for comparison to depression (MVR, DC) and PTSD (TBA, NHC) results. Table cells marked N/A indicate unavailable metrics from previous studies.