DataTerminology.comNEW
Using historical and current data to predict future values or trends.
DESCRIPTION
Forecasting applies statistical and machine learning methods to estimate future outcomes based on historical patterns and explanatory signals. It underpins planning for demand, supply, staffing, finance, and risk. Techniques range from classical models (ARIMA, exponential smoothing) to machine learning and deep learning. Good practice includes feature engineering, hierarchical reconciliation, and robust evaluation with backtesting. Communicating uncertainty via prediction intervals is essential for decision-making.
ALTERNATIVE DEFINITIONS
Wikipedia
Forecasting is the process of making predictions of the future based on past and present data and most commonly by analysis of trends.
Highlights use of past and present data and trend analysis.
NIST/SEMATECH e-Handbook of Statistical Methods
Forecasting is the process of making statements about events whose actual outcomes (typically) have not yet been observed.
Emphasizes prediction of unobserved future events.
Forecasting: Principles and Practice
Forecasting is about predicting the future as accurately as possible, given all the information available, including historical data and knowledge of any future events that might impact the forecasts.
Stresses using all available information, not just history.
RELATED TERMS
Temporal Data ARIMA Exponential Moving Average Regression Analysis Seasonality Cross-Validation Prophet Prediction Interval
COMPETING TERMS
Nowcasting Backcasting
RELATIONS
KNOWLEDGE AREAS
Data Science AI Mathematics MLOps
TARGET AUDIENCE
Data Scientist AI Engineer MLOps Engineer BI Specialist Data Analyst
EXTERNAL SOURCES
- Wikipedia - Forecasting
- Hyndman & Athanasopoulos - Authoritative forecasting textbook
- Statsmodels - Python time series documentation
- SAS - Enterprise forecasting overview
TOP QUESTIONS
What is a hot topic in Forecasting today?
Hierarchical reconciliation, probabilistic forecasts, and machine learning approaches like gradient boosting and transformers are hot. Combining domain knowledge with ML improves robustness.
Category: Topic
How do you evaluate Forecasting models?
Use backtesting with rolling-origin evaluation and metrics like MAE, RMSE, MAPE, or pinball loss for quantiles. Consider business costs of over/under-forecasting.
Category: Validation
What data is required for Forecasting?
Historical signals at suitable granularity, calendar features, promotions, and covariates help. Ensure stationarity assumptions or use models robust to nonstationarity.
Category: Requirements
What are common mistakes in Forecasting?
Data leakage, ignoring seasonality or holidays, no uncertainty estimates, and not handling intermittent demand can mislead decisions.
Category: Mistakes
How does Forecasting differ from Nowcasting?
Forecasting looks ahead beyond current conditions, while nowcasting estimates current or very near-term values using real-time data. Both complement each other.
Category: Comparison
Why communicate uncertainty in Forecasting?
Decision-makers need ranges, not single points. Prediction intervals and quantiles convey risk and support inventory, staffing, and financial decisions.
Category: BestPractice
When should ML be preferred over classical Forecasting?
When complex nonlinear relationships, many covariates, or cross-series features exist. For short, stable series, classical models are often superior and simpler.
Category: Opinion
How is Forecasting operationalized in MLOps?
Automate retraining, backtesting, monitoring drift, and champion-challenger deployments. Track versions and data with experiment management.
Category: MLOps
Where to start learning Forecasting?
Begin with FPP3 (otexts.com) and Statsmodels tutorials. Practice on public time series datasets and implement backtests early.
Category: GettingStarted