Published October, 2022
Actively maintained.
-
Tabularizing time series data
- Features from the target
- Features from exogenous variables
- Single step forecasting
-
Challenges in feature engineering for time series
- Train-test split
- Pipelines
- Multistep forecasting
- Direct forecasting
- Recursive forecasting
-
Time series decomposition
- Components of a time series: trend and seasonality
- Multiplicative and additive models
- Log transform and Box-Cox
- Moving averages
- LOWESS, STL, and multiseasonal time series decomposition
-
Missing data imputation
- Forward and backward filling
- Linear and spline interpolation
- Seasonal decomposition and interpolation
-
Outliers
- Rolling statistics for outlier detection
- LOWESS for outlier detection
- STL for outlier detection
-
Lag features
- Autoregressive processes
- Lag plots
- ACF, PACF, CCF
- Seasonal lags
- Creating lags with open-source
-
Window features
- Rolling windows
- Expanding windows
- Exponentially weighted windows
- Creating window features with open-source
-
Trend features
- Using time to model linear trend
- Polynomial features of time to model non-linear trend
- Changepoints & piecweise linear trends to model non-linear trend
- Forecasting time series with trend using tree-based models
- Creating trend features with open-source
-
Seasonality features
- Seasonal lags
- Seasonal dummies
- Seasonal decomposition methods
- Fourier terms
- Creating seasonality features with open-source
-
Datetime features
- Extracting features from date and time
- Periodic features
- Calendar events
- Creating datetime features with open-source
-
Categorical Features
- One hot encoding
- Target encoding
- Rolling entropy and rolling majority