Recommended Free Tools
Hospitals should compare specific forecasts on the same admissions target, geography, forecast lead time and real-time data—not assume that AI/ML or traditional epidemiological models are better as a class. The latest CDC FluSight results show that many approaches can beat a simple baseline, but they do not establish a universal winner or show which forecast will work best for an individual hospital.
What the latest public benchmark does—and does not—show
CDC’s FluSight 2025–2026 evaluation, published September 30, 2026, solicited probabilistic forecasts of weekly influenza hospital admissions for the current week through three weeks ahead. Forecasts covered the United States nationally and by state, Puerto Rico and Washington, D.C. The comparison baseline carried forward the previous week’s admissions.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Gordis Epidemiology | $50.69 | Buy on Amazon |
| 2 |
|
Modern Epidemiology | $142.99 | Buy on Amazon |
| 3 |
|
Introduction to Epidemiology | $31.92 | Buy on Amazon |
| 4 |
|
Introduction to Epidemiology | $49.81 | Buy on Amazon |
| 5 |
|
Gordis Epidemiology | $33.06 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
The evaluation included 39 of 53 submitted models from 34 teams. Thirty-three of the 39 included models performed better than the carry-forward baseline. The CDC FluSight ensemble ranked seventh by average relative weighted interval score (WIS), and it was one of 12 models that beat the baseline in every jurisdiction. These are results for individual submissions in one public benchmark, not a controlled finding that one model family outperforms another.
CDC classifies model components from descriptions supplied in model metadata. The categories include statistical, mechanistic, AI/ML and ensemble components, and can overlap. AI/ML descriptors include terms such as neural networks, deep learning, LSTM, random forest, SVM and LightGBM; mechanistic descriptors include SEIR/SIR, compartment, renewal and dynamics models. Those labels help describe what a forecast contains, but do not isolate the effect of “AI” versus “traditional” modeling.
#1 Best Overall
Why a strong season average may hide a consequential miss
Forecast performance can change sharply across epidemic phases. In CDC’s 2024–2025 evaluation, the FluSight ensemble led submitted models on average relative WIS and beat the baseline in each jurisdiction, yet its two-week prediction intervals covered only 6% of observed values across jurisdictions for the January 4, 2025 observation at the first peak. Coverage stabilized later in the season. That peak-period result is a specific observation, not a whole-season estimate.
The 2025–2026 evaluation also reports that the CDC ensemble’s prediction intervals struggled during rapid changes. For capacity planning, a model’s performance around onset, acceleration, peak and decline can matter more than its pooled seasonal average.
Rank #2
Compare forecasts on the decision the hospital needs to make
Before comparing algorithms, make sure candidates answer the same operational question. A prediction of positive tests, influenza-like illness, emergency visits, admissions or total census is not interchangeable with a forecast of weekly influenza admissions. The current FluSight target is weekly admissions; a hospital may need a different denominator or population for staffing, bed or supply decisions.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →| Comparison dimension | What to align or inspect | Why it matters |
|---|---|---|
| Outcome and denominator | Weekly admissions, ED visits, positive tests or census; total counts and any patient group or unit that drives the decision. | A forecast for another outcome cannot directly answer the hospital’s question. |
| Forecast horizon | Current week and each lead time through the point when the hospital can still act. | Useful lead time differs by decision; CDC’s current public benchmark extends through three weeks ahead. |
| Geography | Hospital, catchment area, region, state or national level. | Aggregated performance does not prove accuracy for a local patient flow. |
| Epidemic phase | Onset, acceleration, peak timing and height, decline and unusual waves. | Errors at turning points can be obscured by season-wide averages. |
| Inputs and latency | Local admissions, surveillance feeds, revisions, reporting delays and auxiliary predictors; note when each input becomes available. | Late or revised data can make a retrospective forecast look better than the forecast available to decision-makers at the time. |
| Method and assumptions | Statistical/time-series, mechanistic, AI/ML or hybrid components; training history, assumptions and update method. | A category label alone does not establish fit for the hospital’s data or intended use. |
| Operational usability | Uncertainty communication, update cadence, maintenance, access and connection to a specific staffing, bed or supply decision. | A forecast that cannot be interpreted or delivered in time may not support the action it was intended to inform. |
Score uncertainty as well as the central forecast
Influenza demand is uncertain, so evaluate probabilistic forecasts rather than ranking models only by a point estimate. CDC reports relative WIS against its carry-forward baseline: a value below one indicates better performance than that baseline. Read this alongside prediction-interval coverage, which measures how often the observed outcome falls inside the forecast interval. High coverage and the lowest relative WIS do not always belong to the same model, so neither metric alone is enough.
Rank #3
- Recognized by Book Authority as one of the best Public Health books of all time, Introduction to Epidemiology is a comprehensive, reader-friendly introduction to this exciting field.
- Designed for students with minimal training in the biomedical sciences and statistics, this full-color text emphasizes the application of the basic principles of epidemiology according to person, place, and time factors in order to solve current, often unexpected, and serious public health problems.
- Students will learn how to identify and describe public health problems, formulate research hypotheses, select appropriate research study designs, manage and analyze epidemiologic data, interpret and apply results in preventing and controlling disease and health-related events.
- Use a proper probabilistic score. Compare relative WIS (or a suitable alternative) for each forecast horizon and local unit. Keep the baseline and scoring procedure the same across candidates.
- Check interval coverage. Assess whether stated intervals contain observed admissions at their intended levels, and inspect the result by horizon and epidemic phase.
- Measure operationally relevant point errors. Where the decision depends on a central estimate, report point-error metrics as well as probabilistic scores.
- Score the peak explicitly. Track errors in peak timing and magnitude, and assess onset and decline separately rather than relying on a pooled season result.
A 2019 collaborative assessment of 22 models over seven seasons found that more than half consistently beat a historical seasonal-average baseline for several influenza-like-illness targets and for peak timing or magnitude. It also found reporting delays strongly and negatively associated with forecast accuracy in some regions. That study provides context on data access and timeliness; it was not a contemporary head-to-head test of hospital-admission forecasts. See the multiyear assessment.
Run a fair local comparison
- Specify the decision and target. Name the action (for example, staffing or bed capacity), the forecast outcome and denominator, the local geography, and the lead time at which a forecast can change that action.
- Recreate what was knowable at each forecast date. Use matched historical cutoffs and preserve reporting delays, missing values and data revisions. Do not let a model train or forecast with information that was unavailable at that point in real time.
- Include a simple baseline. Compare every candidate with the same baseline, such as carrying forward the previous week’s admissions, rather than judging a complex model in isolation.
- Score each horizon and phase. Report probabilistic accuracy and interval coverage separately by lead time; add point errors and peak timing or magnitude errors where they map to the decision.
- Test across seasons and local units. Show results by geography and epidemic phase, not only as a pooled average. CDC’s jurisdiction-level results and peak-period interval weaknesses show why aggregate rankings can conceal local or time-specific problems.
- Require a usable model description. Ask providers to disclose inputs, update schedule, assumptions, uncertainty, behavior with missing or revised data, and maintenance requirements.
- Monitor prospectively before relying on it. For high-impact staffing or capacity decisions, run the forecast alongside usual planning, review its performance as new outcomes arrive, and retain a human decision process.
Model outputs should be used for their intended purpose, with limitations understood by the people making decisions. CDC’s 2016 guidance on modeling and public-health decision-making emphasizes dialogue between modelers and decision-makers about those limitations. The published evidence cited here does not establish that deploying one model class improves a particular hospital’s staffing, bed or patient outcomes.
Rank #4
Why a hybrid forecast may be worth testing
A hybrid approach combines empirical forecasting with epidemiological structure; it is a design option to evaluate, not a guaranteed middle ground. A 2025 PNAS study tested “epimodulation,” an epidemiological adjustment applied to five empirical models in retrospective U.S. influenza hospital-admission forecasts from January 2022 to May 2023. The authors reported an average 32.9% improvement in their accuracy measure across the overall period (range 24.2–43.7%) and 43.8% during the December 2022–March 2023 seasonal wave (range 30.2–54.5%), compared with the base versions of those tested models. The reported gains are specific to that study, period, models and comparator; they do not establish a universal advantage over all mechanistic models or predict a gain at any particular hospital. See Gibson and coauthors’ study.
Free tools Windows power users keep installed
One-click scans. No signup required.
Ensembling is another option, but its size should not be treated as a fixed rule. A 2024 CDC Emerging Infectious Diseases study found that more than three forecast models were needed for robust ensemble accuracy across the historical hub datasets it analyzed; that result is not a universal optimum for a hospital’s own data. See Optimizing Disease Outbreak Forecast Ensembles.
Best Value
What a hospital can conclude
Public evaluations show that diverse approaches can outperform a simple baseline, while rankings and uncertainty performance vary by jurisdiction, horizon and epidemic phase. They do not answer which forecast is best for a particular hospital. That requires matched local retrospective evaluation followed by prospective monitoring against the decisions the forecast is meant to support.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
