About Forecasting
How Germwatch predicts short‑term illness trends using real‑time outbreak signals.
Germwatch forecasting provides short‑term projections of how illnesses may behave over the next 7–14 days. Forecasts are currently based on real‑time outbreak signals and recent trend movement. Seasonal modelling is built into Germwatch, but will activate once enough historical data has accumulated to produce reliable baselines.
• Trend analysis – recent movement and acceleration
• Confidence intervals – realistic upper and lower bounds based on trend uncertainty
Seasonal baselines will be added automatically once Germwatch has collected sufficient multi‑year data. Until then, forecasts rely on trend‑based projections designed to be stable, interpretable, and transparent.
Trend Analysis
Recent observed values are analysed using simple linear regression to estimate direction and acceleration.
Seasonal Baselines
Germwatch includes a full seasonality engine capable of generating weekly expected illness levels for every syndrome and region. However, meaningful seasonal baselines require multi‑year data. Germwatch is still new, so seasonal patterns will activate once enough historical data exists.
Confidence Intervals
Forecasts include 95% confidence intervals based on trend uncertainty. These intervals widen naturally when recent activity is noisy or unstable, helping users understand the range of plausible outcomes.
Trend Analysis
Germwatch applies a simple linear regression to recent observed values for each syndrome. This identifies whether activity is rising, falling, or stable, and how quickly it is changing.
The model calculates:
- slope – the rate of change
- intercept – the baseline level
- trend projection – expected next‑week value
Trend projections form the basis of Germwatch forecasting today. Seasonal blending will be added once multi‑year data is available.
Seasonal Baselines
Many illnesses follow predictable seasonal patterns. Respiratory activity typically rises in winter, gastrointestinal activity often peaks in early spring, and febrile syndromes show characteristic fluctuations throughout the year.
Germwatch includes a seasonality model capable of generating weekly expected values for each syndrome and region. This model uses multi‑year data to calculate:
- expected weekly illness levels
- seasonal confidence intervals
- smoothed week‑to‑week behaviour
Because Germwatch is still new, it does not yet have enough historical data to activate seasonal baselines. As data accumulates, seasonal modelling will automatically switch on and begin blending with trend projections.
Confidence Intervals
Germwatch provides 95% confidence intervals for each forecast. These intervals are currently based on trend uncertainty, using an overdispersed Poisson model to reflect real‑world variability in illness reporting. As seasonal baselines become available, confidence intervals will incorporate both trend and seasonal variance.
Confidence intervals help users understand:
- the range of likely outcomes
- forecast stability
- uncertainty during rapid changes
Forecast Confidence
Germwatch assigns a confidence score (0–5) based on the strength of the trend. Strong, consistent trends produce higher confidence, while flat or noisy trends produce lower confidence.
- 2 – minimal trend
- 3 – mild trend
- 4 – clear trend
- 5 – strong trend
Germwatch forecasting uses real‑time trends and statistical uncertainty to produce clear, short‑term illness projections. Seasonal modelling is built in and will activate automatically once enough historical data exists.