/feature-engineering-for-time-series-forecasting

Primary LanguageJupyter NotebookOtherNOASSERTION

Feature Engineering for Time Series Forecasting - Code Repository

PythonVersion License https://github.com/trainindata/feature-engineering-for-time-series-forecasting/blob/master/LICENSE Sponsorship https://www.trainindata.com/

Published October, 2022

Actively maintained.

Links

Table of Contents

  1. Tabularizing time series data

    1. Features from the target
    2. Features from exogenous variables
    3. Single step forecasting

  2. Challenges in feature engineering for time series

    1. Train-test split
    2. Pipelines
    3. Multistep forecasting
    4. Direct forecasting
    5. Recursive forecasting

  3. Time series decomposition

    1. Components of a time series: trend and seasonality
    2. Multiplicative and additive models
    3. Log transform and Box-Cox
    4. Moving averages
    5. LOWESS, STL, and multiseasonal time series decomposition

  4. Missing data imputation

    1. Forward and backward filling
    2. Linear and spline interpolation
    3. Seasonal decomposition and interpolation

  5. Outliers

    1. Rolling statistics for outlier detection
    2. LOWESS for outlier detection
    3. STL for outlier detection

  6. Lag features

    1. Autoregressive processes
    2. Lag plots
    3. ACF, PACF, CCF
    4. Seasonal lags
    5. Creating lags with open-source

  7. Window features

    1. Rolling windows
    2. Expanding windows
    3. Exponentially weighted windows
    4. Creating window features with open-source

  8. Trend and seasonality features

    1. Trend: regression over time
    2. Seasonal dummies
    3. Fourier terms

  9. Datetime features

    1. Extracting features from date and time
    2. Periodic features
    3. Calendar events
    4. Creating datetime features with open-source

  10. Categorical Features

    1. One hot encoding
    2. Target encoding
    3. Rolling entropy and rolling majority