fraunhoferportugal/tsfel

What it solves

TSFEL simplifies the process of extracting meaningful characteristics (features) from time series data, which is a critical step for preparing data for machine learning models. Instead of manually implementing dozens of different mathematical formulas, users can access a centralized library of over 65 distinct features across multiple domains.

How it works

The library provides a high-level API that allows users to define a configuration for the features they want to extract (e.g., by domain) and then apply that configuration to their time series data. It supports four primary domains of analysis:

• Statistical: Basic metrics like mean, variance, and entropy.
• Temporal: Analysis of time-based patterns, such as zero crossing rates and autocorrelation. -- Spectral: Frequency-based analysis using tools like Fast Fourier Transforms (FFT) and Wavelets.
• Fractal: Analysis of complex patterns over longer signals, such as the Hurst exponent.

Who it’s for

Data scientists and researchers working with time series analysis, such as those analyzing ECG signals or sensor data for Human Activity Recognition (HAR).

Highlights

• Comprehensive Feature Set: Access to over 65 features across statistical, temporal, spectral, and fractal domains.
• Reproducible Pipelines: Configuration files can be stored to ensure the same extraction process is used across different datasets.
• Computational Cost Estimation: Ability to estimate the time required for feature extraction in advance.
• Extensible: Users can easily add their own custom feature extraction methods to the library.