What Confidence Means
Model output probability is not certainty. Here is what the confidence score represents - and what it does not.
Confidence as Model Output Probability
Each Seismikon prediction carries a confidence score between 0 and 1 (or 0–100%). This score reflects the LSTM model's internal output probability for that particular prediction - broadly, how strongly the model's pattern-matching favoured this forecast over alternatives, given the current state of the seismic system.
A higher confidence score means the model found stronger signal in the current data for this specific prediction. It does not mean the event is certain to occur.
What Confidence Does Not Mean
Several common misreading of confidence scores should be avoided:
- Confidence ≠ probability of occurrence. A 90% confidence score does not mean there is a 90% chance the earthquake will happen. The model score has not been calibrated to that interpretation.
- Confidence ≠ precision. A high-confidence prediction may still fall outside tolerance if the seismic system behaves unexpectedly.
- Low confidence ≠ no value. Lower-confidence predictions still contribute to the population of forecasts evaluated for skill. A system that only issues very high-confidence predictions and abstains otherwise is harder to evaluate.
Calibration and Future Work
For confidence scores to be interpretable as probabilities, the model's outputs need to be calibrated - meaning that across all predictions issued at 70% confidence, roughly 70% should be true positives. Calibration analysis requires a large population of predictions and is on the roadmap as more prospective data accumulates. Current confidence scores should be treated as ordinal rankings (higher = stronger signal) rather than absolute probabilities.