All Tools
T
DataFreeOpen Source
TSFRESH
Automatic time-series feature extraction
MIT
ABOUT
Time-series feature engineering requires extensive domain knowledge in signal processing, statistics, and spectral analysis to extract meaningful patterns — practitioners must manually implement lag features, rolling statistics, Fourier transforms, and peak detection for each use case. tsfresh automates this by extracting 794 pre-defined features from each time series (mean, variance, FFT coefficients, autocorrelation, entropy, wavelet transforms), tests them for significance against the target variable, and filters irrelevant features — enabling non-experts to build high-quality time-series features in minutes.
INSTALL
pip install tsfreshINTEGRATION GUIDE
1. Extract hundreds of statistical and spectral features from sensor time-series data for predictive maintenance, anomaly detection, and equipment failure prediction
2. Automatically filter relevant time-series features for forecasting workloads (demand forecasting, energy load prediction) using the built-in hypothesis-testing feature selector
3. Generate scikit-learn-compatible feature matrices from large collections of time series for classification and regression on sequential data
TAGS
time-seriesfeature-extractionfeature-engineeringpythonsignal-processingmachine-learningdata-preparation