DataTerminology.comNEW
Modeling historical time series to predict future values, including uncertainty estimates over specified horizons.
DESCRIPTION
Time series forecasting uses past observations to project future values. Classical approaches include ARIMA and exponential smoothing; modern methods add machine learning and deep learning. Seasonality, trend, and external regressors enhance accuracy. Proper evaluation relies on time-aware backtesting and metrics like MAPE, MAE, or pinball loss for quantiles. Prediction intervals communicate uncertainty. Robust pipelines manage feature lags, holidays, and data leakage. Forecasts support capacity planning, inventory, pricing, and finance.
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.
General definition of forecasting applicable to time series.
Hyndman & Athanasopoulos
Forecasting is about predicting the future as accurately as possible, given all the information available, including historical data and knowledge of any future events.
Emphasizes using all available information and accuracy focus.
NIST Dictionary of Algorithms and Data Structures
Forecasting is the estimation of the value of a variable at some future time.
Concise, variable-focused definition from a government authority.
RELATED TERMS
Temporal Data ARIMA Exponential Moving Average Prophet Seasonality Stationarity Prediction Interval Forecast Horizon Backtesting Cross-Validation
COMPETING TERMS
Judgmental Forecasting
RELATIONS
- Broader: Forecasting
KNOWLEDGE AREAS
TARGET AUDIENCE
Data Scientist Data Analyst AI Engineer BI Specialist AI Product Manager
EXTERNAL SOURCES
- Wikipedia - Forecasting
- Hyndman & Athanasopoulos - Open forecasting textbook
- statsmodels - Time series analysis API
- Prophet - Forecasting with Prophet
TOP QUESTIONS
What is the first step in Time Series Forecasting?
Understand data generation and business context. Explore seasonality, trend, and anomalies, then define horizon, granularity, and accuracy targets before selecting models.
Category: HowTo
How do you prevent leakage in Time Series Forecasting?
Strictly maintain temporal order in feature creation and validation. Use rolling-origin evaluation and avoid using future information in training.
Category: BestPractices
Which metrics are best for Time Series Forecasting?
MAPE, MAE, RMSE, and quantile pinball loss for probabilistic forecasts. Metric choice should reflect business loss functions.
Category: Validation
When use ARIMA vs. machine learning models?
ARIMA works well for univariate, stable patterns. Machine learning can exploit exogenous variables and complex nonlinearities, especially at scale.
Category: Comparison
How to incorporate seasonality in Time Series Forecasting?
Use seasonal ARIMA/ETS, seasonal dummies, Fourier terms, or learned seasonal components in models like Prophet or neural nets.
Category: HowTo
What is a common mistake in Time Series Forecasting?
Not recalibrating models as regimes change. Concept drift degrades accuracy, requiring retraining and monitoring.
Category: Mistakes
Why are prediction intervals crucial in Time Series Forecasting?
They quantify uncertainty, enabling risk-aware decisions and safety stock planning. Point forecasts can be misleading without uncertainty.
Category: Relevancy
How do holidays and events affect Time Series Forecasting?
Special events cause predictable deviations. Explicit features or holiday calendars improve accuracy markedly.
Category: Example
What is backtesting in Time Series Forecasting?
Training and evaluating forecasts on historical rolling windows to estimate out-of-sample performance. It mimics future deployment conditions.
Category: Definition