FLIRT

FLIRT standardizes preprocessing and feature generation for physiological data from commercial wearables such as smartwatches and fitness trackers to produce reproducible, machine-learning–ready feature sets for health and behavioral studies.

Key Features:

• Robust Preprocessing: Employs particle filters and machine learning-based artifact detection to manage noise, missing data, and artifacts in wearable sensor signals.
• Standardized Feature Generation: Uses a sliding-window approach to compute feature vectors with over 100 dimensions for downstream machine learning applications.
• Integrated File Format Handling: Supports common file formats from popular wearables, including Empatica E4, to facilitate data integration.
• Configurable Algorithms: Allows customization of preprocessing and feature-generation algorithms to accommodate specific research needs.

Scientific Applications:

• Stress detection (WESAD evaluation): Applied to stress detection and evaluated on the WESAD dataset using an Empatica E4 wearable, demonstrating superior performance compared to existing baselines.
• Activity and behavioral monitoring: Enables prediction of physical activity patterns and other behavioral outcomes through standardized feature extraction for machine learning models.

Methodology:

Preprocessing uses particle filters and machine learning-based artifact detection; features are computed with a sliding-window approach producing >100-dimensional vectors; supports parsing common wearable file formats including Empatica E4.