Published October 12, 2026 | Version v1

Unsupervised Learning Reveals Recurring Behavioral Patterns During Real-World Wayfinding (Code base)

  • 1. ROR icon TU Wien

Description

This codebase accompanies the paper “Unsupervised Learning Reveals Recurring Behavioral Patterns During Real-World Wayfinding” and provides the Python scripts used to analyze gaze and head-movement data collected during real-world pedestrian navigation. The study uses the same dataset as the one publicly available at https://doi.org/10.48436/m2ha4-t1v92.

The main analysis applies Deep Embedded Clustering (DEC) to identify recurring behavioral patterns without imposing predefined wayfinding stages or behavioral labels. The pipeline preprocesses and normalizes the sensor recordings, divides the continuous signals into overlapping temporal windows, and uses a deep autoencoder to learn compact representations of gaze and head-movement behavior. Clustering is then performed on these learned representations, with the number of clusters selected using the Silhouette score. Different sensor combinations are evaluated to investigate the behavioral structures captured by gaze and inertial measurements.

The codebase also includes additional analyses to examine the recurrence, consistency, and robustness of the identified patterns. These include comparisons between familiar and unfamiliar navigation, sensitivity analyses using different temporal window sizes, clustering stability assessments across random seeds using the Adjusted Rand Index (ARI), and summaries of participant and route contributions to the analysis.

Files

requirements.txt

Files (67.7 KiB)

NameSize
md5:03e4799d38902d7c6ee6769cb0b1b864
13.9 KiBPreview Download
md5:f13c86eaf4b2e5c7eac784893dc3398e
1.3 KiBPreview Download
md5:8a600d784d0844f9dfe6cee96b51caf4
6.4 KiBPreview Download
md5:de8ad48e831daec3203123a43abaf0a9
71 BytesPreview Download
md5:ac8b40b3f64b1b06279d5f5e9459dc89
1.7 KiBPreview Download
md5:633b6a161e9cc25816ba24b2182d219f
44.3 KiBPreview Download

Additional details

Related works

Is supplement to
Dataset: 10.48436/m2ha4-t1v92 (DOI)