Using dynamic time warping self-organizing maps to characterize diurnal patterns in environmental exposures

Area of Science:

• Environmental Health
• Data Science
• Computational Biology

Background:

• High-frequency environmental exposure data are increasingly available.
• Traditional summary statistics may oversimplify complex exposure patterns.
• Understanding detailed time-series exposures is crucial for health impact assessment.

Purpose of the Study:

• To introduce a novel algorithm, Dynamic Time Warping Self-Organizing Map (DTW-SOM), for unsupervised pattern discovery in time-series data.
• To analyze high-frequency indoor and outdoor environmental exposures (temperature, PM2.5).
• To investigate potential relationships between detailed environmental patterns and asthma outcomes.

Main Methods:

• Developed and applied a novel Dynamic Time Warping Self-Organizing Map (DTW-SOM) algorithm.
• Utilized Dynamic Time Warping (DTW) as both a similarity measure and training guide for the neural network.
• Analyzed time-series data from a panel study of 10 asthma patients, monitoring residential temperature and PM2.5.

Main Results:

• DTW-SOM demonstrated superior performance over traditional SOM algorithms, yielding fewer errors and more refined diurnal patterns.
• The algorithm successfully identified seasonal variations in outdoor temperature patterns.
• DTW-SOM revealed diurnal patterns in PM2.5, suggesting potential links to daily asthma exacerbations.

Conclusions:

• DTW-SOM is an innovative feature engineering technique for high-resolution sensor data.
• This method effectively identifies typical diurnal, hourly, or monthly patterns in environmental exposures.
• DTW-SOM offers novel insights into the health effects of environmental exposures, particularly for conditions like asthma.