Extended Data Fig. 7: Distribution of performance measures per forecasting model.
From: Hippocampal network activity forecasts epileptic seizures
Distributions of FC-based model performance (blue) and IEA cycle-based model performance (yellow) across participants (n = 15; box plots show the maximum and minimum values (whiskers), median (center line) and the 25th to 75th percentiles (box limits); violin shading shows kernel density) for (a) precision, (b) recall, and (c) F1 measures. There is no significant difference in precision, recall, or F1 score between the two models (two-sided paired t-test, P >= 0.05, Bonferroni corrected), suggesting that an estimate of momentary FC can forecast seizures as well as the benchmark model that uses multidien IEA phase measured over a timescale of 4–45 days as a predictor. (d) Receiver operating characteristic curve for FC-based models (blue), IEA cycle-based models (yellow), and shuffled baseline (gray) across participants (n = 15). There is no significant difference in the rate of false positives compared to the rate of true positives between the FC-based model and the cycle-based model (two-sided paired t-test, P >= 0.05, Bonferroni corrected). Thick lines, mean; shading, SEM; dashed line, unity line.
