Dynamic Elderly Care Resource Allocation using SHBM-Tuned Deep Q-Networks (SHBM-RDQN) Hua

Abstract

The growing demand for elderly care services necessitates intelligent, efficient, and adaptive resource management strategies. This research presents a dynamic allocation strategy for optimizing elderly care resources using reinforcement learning (RL), addressing the complexity of nursing home resource management. This research proposes a Dynamic Honeybees Mating–tuned Resource-based Deep Q-Network (DHBM-RDQN) for elderly care resource allocation. The approach models the care environment as a Markov Decision Process (MDP), where states capture patient acuity levels, staff availability, daily admissions, and care workload, while actions correspond to dynamic allocation of nursing staff and support resources. The system uses a comprehensive Elderly Care Staffing & Quality Dataset, comprising time-series records at daily or shift-level granularity from multiple long-term care facilities. Preprocessing includes Z-score normalization and missing value imputation, while feature extraction applies Independent Component Analysis (ICA) and Discrete Wavelet Transform (DWT) to capture latent health patterns and temporal fluctuations. The DHBM-RDQN employs a Deep Q-Network with two fully connected layers (256 and 128 neurons) and ReLU activations, trained using the Adam optimizer. The Honeybee Mating optimization layer dynamically tunes learning rate, exploration parameters, and reward weights to prevent premature convergence. Experimental evaluation were implemented in python. Results show 96.5% accuracy, 90.5% resource efficiency, 1.0 s response time, and an adaptability score of 0.925, demonstrating robust adaptability under fluctuating patient demand and staffing. This research introduces a novel RL framework combining deep learning and bio-inspired optimization to achieve superior performance, rapid decision-making, and improved care quality in dynamic long-term elderly care environments.