Privacy-Preserving Multisensor Edge Intelligence Using Split Dual-LSTM Architecture for Real-Time Elderly Care

Abstract

The elderly population is rapidly growing and needs reliable and dependable healthcare monitoring systems that are privacy-protecting. Traditional cloud-centric healthcare architectures introduce latency and data-exposure vulnerabilities that make real-time response challenging. This article presents a privacy-preserving multisensor edge intelligence (PSEI) framework that integrates inertial measurement unit (IMU) and received signal strength indicator (RSSI) data by using a split dual-long short-term memory (LSTM) model. The model divides the tasks of recognizing activities and finding their locations between wearable sensors and an edge server. This enables privacy-sensitive raw signals to stay local while computationally intensive inference tasks are handled at the edge. A software-defined networking (SDN) layer dynamically optimizes routing and congestion control to keep ultrareliable low-latency communication (URLLC). The experimental results showed that 86%–96% of packets were sent within 15 ms and 80%–90% of the models were accurate. There was a drop of more than 50% in delayed packets when there were 15 or more hosts and heavy traffic. This represents approximately a 25% improvement in latency compliance over similar edge frameworks. The experimental results demonstrate network-level readiness and architectural suitability for Tactile-Internet-enabled healthcare scenarios, achieving URLLC-grade latency and reliability under dense Internet of Medical Things (IoMT) traffic conditions. PSEI is a strong base for the next generation of IoMT-based health-monitoring and elderly care systems because it combines sensor-level privacy, latency resilience, and scalability.