Abstract
Accurate classification of Activities of Daily Living (ADLs) is critical for monitoring functional decline in individuals with Alzheimer's disease and related dementias (ADRD). This study evaluates ambient sensor-based ADL recognition using data from in two real-world dyads involving individuals with early-stage ADRD and their caregivers. The sensing setup included motion, temperature, humidity, luminance, and door sensors deployed across key locations. We compare stacked Long Short-Term Memory (LSTM) networks with Transformer models that leverage multi-head self-attention to capture temporal dependencies in sensor data. Model performance was evaluated using micro-averaged AUC-ROC and F1 score to account for class imbalance. Results show that Transformer models outperform LSTMs, achieving peak AUCs of 0.93 and 0.90 across the two homes. These findings show consistent performance gains of attention-based models for real-world ambient sensing and highlight their potential to support early detection of functional decline, enabling timely interventions and reducing caregiver burden.
| Original language | English (US) |
|---|---|
| Title of host publication | Proceedings - 2026 IEEE/ACM Conference on Connected Health |
| Subtitle of host publication | Applications, Systems and Engineering Technologies, CHASE 2026 |
| Publisher | Institute of Electrical and Electronics Engineers Inc. |
| Pages | 579-580 |
| Number of pages | 2 |
| ISBN (Electronic) | 9798319507525 |
| DOIs | |
| State | Published - 2026 |
| Event | 2026 IEEE/ACM Conference on Connected Health: Applications, Systems and Engineering Technologies, CHASE 2026 - Pittsburgh, United States Duration: Aug 4 2026 → Aug 6 2026 |
Publication series
| Name | Proceedings - 2026 IEEE/ACM Conference on Connected Health: Applications, Systems and Engineering Technologies, CHASE 2026 |
|---|
Conference
| Conference | 2026 IEEE/ACM Conference on Connected Health: Applications, Systems and Engineering Technologies, CHASE 2026 |
|---|---|
| Country/Territory | United States |
| City | Pittsburgh |
| Period | 8/4/26 → 8/6/26 |
Bibliographical note
Publisher Copyright:© 2026 IEEE.
Keywords
- ambient sensing
- deep learning
- early-stage adrd
- time series data
Fingerprint
Dive into the research topics of 'Ambient Sensor-Based Activities of Daily Living ADLs Classification in Early Alzheimer's Disease: A Real-World Evaluation'. Together they form a unique fingerprint.Cite this
- APA
- Standard
- Harvard
- Vancouver
- Author
- BIBTEX
- RIS