07 Chapter

Time-Series Forecasting

Predict future values from trends, seasonality, and lagged signals.

Time-series forecasting predicts future values of ordered sequences from trends, seasonality, and lagged signals. The methods below span classical statistical models, feature-based gradient boosting, and modern neural sequence models.

  • For business forecasting, start with ETS/SARIMA.
  • Then try gradient boosting with lag features, then deep learning if you have lots of data.
#AlgorithmBest forCommon fields
1ARIMA / SARIMA Classical univariate forecasting
  • Economics
  • demand
  • energy
  • finance
2Exponential Smoothing / ETS / Holt-Winters Trend and seasonality
  • Retail
  • inventory
  • operations
3Prophet-style additive models Business forecasting with seasonality/holidays
  • Sales
  • marketing
  • web traffic
4Gradient Boosting for Time Series Feature-based forecasting
  • Demand forecasting
  • pricing
  • operations
5LSTM / GRU Sequence-heavy forecasting
  • Sensors
  • energy
  • finance
6Temporal CNN / TCN Long sequence forecasting
  • Industrial IoT
  • signals
7Transformer Forecasting Models Large-scale multivariate forecasting
  • Retail
  • logistics
  • cloud metrics
8State Space Models / Kalman Filter Dynamic systems
  • Robotics
  • navigation
  • economics