Prediction of basic physiological data for the elderly in community care driven by GNN-LSTM
Accurate prediction of physiological changes is important for continuous health monitoring of elderly individuals in community-care. However, multivariate physiological time-series data exhibit complex temporal dependencies and heterogeneous characteristics, limiting the performance of conventional single-model approaches. This study proposes a graph neural network and long short-term memory network (GNN-LSTM) framework for one-step-ahead prediction of elderly physiological measurements. The multivariate observations are represented as directed temporal chain graphs, where physiological variables are encoded as node attributes. The GNN captures local temporal information, while the LSTM models sequential dependencies to generate future predictions. A leakage-free evaluation strategy is adopted by performing data partitioning before preprocessing and sliding-window construction.Experimental results show that the proposed model achieves R 2 values of 0.936, 0.864, and 0.988 for blood pressure, heart rate, and respiratory-related measurement prediction, respectively, with Mean Absolute Error of 2.468 mmHg, 0.342 bpm, and 1.243 a.u. The proposed framework demonstrates the effectiveness of combining temporal graph representation and recurrent modeling for multivariate physiological signal prediction in elderly community-care applications.
Authors
- Weifu Kong
- Deyan Xu
- Boren Xue (ORCID: https://orcid.org/0009-0006-5697-3970)
- Zhibo Bao
- Guanfeng Du
- Shengxian Cao
Institutions
- Northeast Electric Power University (CN)
- Jilin University of Chemical Technology (CN)
Publication Details
- Journal
- Array
- Published
- 2026-11-20
- DOI
- https://doi.org/10.1016/j.array.2026.101475
- Primary Topic
- Non-Invasive Vital Sign Monitoring
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- People's Government of Jilin Province