Skill Guide

Time-series forecasting for disease incidence and mortality trends

The application of statistical and machine learning models to sequential epidemiological data (cases, deaths, rates) to predict future disease burden and identify underlying trends, seasonality, and anomalies.

This skill enables proactive public health planning, resource allocation (e.g., vaccines, ICU beds), and early outbreak detection. It directly reduces healthcare costs, improves population health outcomes, and supports evidence-based policy-making.

1 Careers

1 Categories

9.0 Avg Demand

25% Avg AI Risk

How to Learn Time-series forecasting for disease incidence and mortality trends

Master foundational time-series concepts: stationarity, autocorrelation (ACF/PACF), decomposition (trend, seasonality, residual). Understand key epidemiological metrics (incidence rate, mortality rate, case fatality ratio). Practice with clean, publicly available datasets (e.g., WHO Weekly Epidemiological Record, CDC WONDER) using basic plotting and moving averages.

Implement classical statistical models (ARIMA/SARIMA, Exponential Smoothing) and validate with train/test splits and metrics like MAE, RMSE, MAPE. Handle real-world data issues: missing values, reporting delays, structural breaks from interventions (lockdowns, vaccines). Explore basic regression with epidemiological covariates (mobility data, testing rates).

Architect ensemble systems combining statistical, ML (Prophet, gradient boosting), and deep learning (LSTM, Temporal Fusion Transformer) models. Integrate non-traditional data streams (syndromic surveillance, wastewater analysis). Design robust forecasting pipelines for operational deployment, incorporating uncertainty quantification (prediction intervals) and model explainability (SHAP) for stakeholder communication. Mentor junior analysts and align forecasts with strategic planning cycles.

Practice Projects

Beginner

Project

Weekly Influenza-Like Illness (ILI) Forecast for a Single Region

Scenario

You are a junior analyst at a state health department tasked with forecasting weekly ILI visits for the upcoming 4 weeks to guide clinic staffing.

How to Execute

1. Acquire and clean 3-5 years of historical ILI% data from CDC FluView. 2. Perform decomposition to visualize trend and strong annual seasonality. 3. Fit a simple SARIMA model using auto_arima or similar. 4. Generate forecasts and evaluate using a rolling-window backtest, presenting point estimates and 95% prediction intervals.

Intermediate

Project

COVID-19 Hospitalization Forecast with Intervention Adjustments

Scenario

Lead the analysis to predict regional hospitalizations 3 weeks out, a period that spans the scheduled lifting of a public mask mandate.

How to Execute

1. Gather daily hospitalization data and append a binary or time-varying feature for mask mandate intensity. 2. Engineer features like mobility indices (Google) and vaccination coverage. 3. Compare a SARIMAX model (with covariates) against a gradient boosting model (XGBoost). 4. Conduct scenario analysis: run forecasts under different assumptions of post-mandate behavior. Document model selection rationale and uncertainty for a public briefing.

Advanced

Project

Multi-Disease Burden Forecasting System for National Resource Planning

Scenario

As a lead data scientist, design and implement a forecasting system to simultaneously predict incidences of influenza, RSV, and a novel respiratory pathogen for the national stockpile committee.

How to Execute

1. Architect a modular pipeline ingesting disparate data (lab reports, ER visits, genomic surveillance, Google Trends). 2. Implement a hierarchical or multi-task model structure to share information across diseases. 3. Build a robust MLOps framework for automated model retraining, monitoring for concept drift, and probabilistic forecast reconciliation (ensuring forecasts of individual diseases don't sum to >100% of respiratory capacity). 4. Create an executive dashboard presenting scenario-based forecasts with clear risk bands for committee decision-making.

Tools & Frameworks

Software & Platforms

Python (statsmodels, Prophet, scikit-learn, PyTorch Forecasting, sktime)R (forecast, fable, tseries)Databricks / Apache Spark for scalable data pipelinesVisualization: Tableau, Power BI, Plotly Dash

Use Python/R for model development and prototyping. Platforms like Databricks are critical for handling large-scale, streaming epidemiological data. Visualization tools are for stakeholder communication and operational dashboards.

Statistical & ML Frameworks

ARIMA/SARIMA/SARIMAX familyExponential Smoothing State Space Models (ETS)Prophet (for strong seasonality and holiday effects)Temporal Fusion Transformer (TFT) for interpretable deep learning forecasts

SARIMA is the benchmark for univariate series with seasonality. Prophet handles multiple seasonalities and missing data well. TFT is state-of-the-art for multi-horizon forecasting with covariates and built-in interpretability.

Mental Models & Methodologies

Epidemiological Modeling (SIR/SEIR as covariate generators)Backtesting & Rolling Window ValidationEnsemble Modeling & Forecast CombinationBayesian Structural Time Series (BSTS) for uncertainty quantification

Use SEIR insights to engineer features, not necessarily for direct forecasting. Rigorous backtesting prevents overfitting to recent trends. Ensembles improve robustness. BSTS provides a principled probabilistic framework.

Interview Questions

Answer Strategy

The question tests the ability to handle non-stationarity, incorporate intervention variables, and manage data quality. Strategy: Detail a step-by-step, practical approach focusing on data cleaning, feature engineering, and model selection. Sample Answer: "First, I would address reporting lags with a nowcasting model or use a smoothing filter like a 7-day rolling average. For modeling, I would use a SARIMAX model, incorporating vaccine coverage (% fully vaccinated) as an exogenous variable. I'd include time dummies to account for reporting policy changes. The forecast would be generated iteratively, and I would heavily emphasize the prediction intervals to convey uncertainty during this transitional period to stakeholders."

Answer Strategy

This behavioral question tests humility, problem-solving, and a commitment to rigorous validation. Strategy: Use the STAR method, focusing on the root cause (e.g., ignoring a structural break) and the process improvement you implemented. Sample Answer: "Situation: During the Delta wave, my flu forecast for a winter season was off by 40% because the model couldn't capture the behavioral shift from mask-wearing fatigue. Task: The error led to a temporary staff shortage in sentinel clinics. Action: I led a post-mortem, identified the need for real-time mobility data as a covariate, and implemented a changepoint detection algorithm. Result: The next iteration's MAPE dropped to under 10%, and we formalized the inclusion of behavioral data sources in our standard pipeline."