ML Signal Detection
Train a random forest on FAERS contingency data to detect drug safety signals. The model learns from labeled signal/noise pairs and predicts new drug-event combinations. All computation runs locally.
Step 1: Enter FAERS Contingency Data
Each row is a drug-event pair with its 2x2 contingency table counts from real FAERS data. Label known signals and noise for training.
| Drug | Event | a (D+E+) | b (D+E-) | c (D-E+) | d (D-E-) | Label | |
|---|---|---|---|---|---|---|---|
How It Works
Each drug-event pair gets 12 features: PRR, ROR, IC, EBGM (disproportionality), plus reporter mix, outcome severity, temporal patterns, and reporting velocity.
100 decision trees trained with bootstrap sampling and feature subsampling. Each tree votes — majority wins. More robust than any single statistical threshold.
New pairs scored by probability (0-100%). The model captures non-linear interactions between features that simple PRR thresholds miss — like low PRR but high seriousness.