Article: Predicting elderly care needs via spatio-temporal transformer and reinforcement learning Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.79 pp.20 - 45 Abstract: As global population aging intensifies, precise prediction of elderly care needs is of vital importance for the optimal allocation of medical resources and personalised intervention. This paper proposes a hybrid framework that integrates spatial-temporal transformer and deep reinforcement learning to predict multiple care needs using longitudinal health and social data. This framework models the long-term dependencies of health trajectories through spatial-temporal Transformer and cross-modal interactions among clinical, functional, and social factors. It also incorporates a reinforcement learning component to optimise the sequential decision-making process of care need evolution. Experiments on two datasets show that the proposed method achieves significant improvements over nine baseline models and four evaluation metrics. Specifically, the AUC increases by 3.2% to 7.8%, and the F1-score increases by 4.1% to 9.3%. This research provides a new methodological tool and empirical basis for elderly care planning. Inderscience Publishers - linking academia, business and industry through research

Title: Predicting elderly care needs via spatio-temporal transformer and reinforcement learning

Authors: Jinrong Tang

Addresses: Medical School, Changsha Social Work College, Changsha, 410004, China

Abstract: As global population aging intensifies, precise prediction of elderly care needs is of vital importance for the optimal allocation of medical resources and personalised intervention. This paper proposes a hybrid framework that integrates spatial-temporal transformer and deep reinforcement learning to predict multiple care needs using longitudinal health and social data. This framework models the long-term dependencies of health trajectories through spatial-temporal Transformer and cross-modal interactions among clinical, functional, and social factors. It also incorporates a reinforcement learning component to optimise the sequential decision-making process of care need evolution. Experiments on two datasets show that the proposed method achieves significant improvements over nine baseline models and four evaluation metrics. Specifically, the AUC increases by 3.2% to 7.8%, and the F1-score increases by 4.1% to 9.3%. This research provides a new methodological tool and empirical basis for elderly care planning.

Keywords: care needs for the elderly; spatiotemporal transformer; deep reinforcement learning.

DOI: 10.1504/IJICT.2026.155057

International Journal of Information and Communication Technology, 2026 Vol.27 No.79, pp.20 - 45

Received: 13 Mar 2026
Accepted: 03 May 2026
Published online: 24 Jul 2026 *

Title: Predicting elderly care needs via spatio-temporal transformer and reinforcement learning

Keep up-to-date