Artificial Intelligence-Driven Enhancement of Sports Wellness in the Elderly via an Adaptive Transfer Learning-Based Backpropagation Neural Network

In this study, a dynamic adaptive decision-making system for sports wellness is constructed to meet the requirements of precision and safety of exercise interventions for the elderly. By integrating multimodal physiological data with sports behavior features, an Adaptive Transfer Learning-Based Backpropagation Neural Network (ATL-BPNN) fused with an attention mechanism is proposed. The model improves the sensitivity to heart rate variability, blood oxygen saturation, and electromyographic signals by using an adaptive weight distribution mechanism. Meanwhile, this model overcomes training difficulties caused by sparse health data among the elderly by combining a transfer learning strategy. Experimental results show that the accuracy of ATL-BPNN in exercise risk prediction reaches 95.2%, which is 12.7% higher than the traditional model; the posture recognition error is reduced to 0.87°. The system achieves a decision latency of less than 50 milliseconds under an edge computing architecture. The practical application proves that this scheme can reduce the incidence of sports injuries by 31.5% and significantly improve the exercise safety environment for the elderly.