WISE-HAR: A Generalizable Ensemble Deep Learning Framework for WiFi-Based Human Activity Recognition

Researchers present WISE-HAR, an ensemble deep learning framework that recognizes human activities using WiFi signals with 94.87% accuracy. The approach combines five CNN architectures with aggressive data augmentation and demonstrates strong cross-scenario generalization, positioning WiFi-based activity recognition as a practical, privacy-preserving alternative to camera and wearable-based systems.

WISE-HAR addresses a growing demand for non-intrusive human activity recognition in smart environments by leveraging WiFi signals rather than cameras or wearables. The research tackles three fundamental challenges that have historically limited WiFi-based HAR deployment: performance inconsistency through ensemble learning, dataset scarcity through data augmentation, and real-world variability through rigorous cross-scenario testing. The framework's ability to maintain 92.5% accuracy when switching from Line-of-Sight to Non-Line-of-Sight conditions, and 92.8% accuracy across different antenna types, demonstrates practical robustness that matters for deployment.

This work emerges within a broader trend toward ambient intelligence and privacy-by-design computing architectures. Smart home technology, healthcare monitoring, and security systems increasingly face regulatory pressure regarding data collection; WiFi-based approaches avoid recording images or requiring personal devices, addressing both privacy concerns and user friction. The dramatic improvement of Random Forest performance from 60% to 95% through augmentation illustrates how modern machine learning techniques can overcome traditional limitations of small datasets.

For the IoT and smart building industries, this research validates a pathway to cost-effective, scalable activity monitoring without the privacy liability of visual systems. Developers can implement WiFi-HAR using existing infrastructure, reducing deployment friction. The 1-2% accuracy degradation under real-world conditions suggests the technology approaches practical viability for commercial applications in healthcare facilities, assisted living environments, and residential spaces where privacy remains paramount.

• →Ensemble learning combining five CNN architectures achieves 94.87% activity recognition accuracy on WiFi spectrograms
• →Cross-scenario testing demonstrates strong generalization with minimal accuracy loss when switching between different hardware and environmental conditions
• →Aggressive data augmentation improved traditional machine learning performance by 58%, addressing dataset size limitations
• →WiFi-based HAR offers privacy-preserving alternative to camera systems while avoiding user compliance issues of wearables
• →Framework recognition of three activity classes (no presence, walking, walking with arm-waving) covers fundamental use cases for smart home and healthcare monitoring