Multi-Modal Deep Learning for Non-Invasive Elder Safety Monitoring and Hybrid Architectures Using Gas Sensor Arrays and Binary Positional IoT Data

Abstract

The global demographic shift toward aging populations necessitates innovative, privacy-preserving solutions for continuous elder safety monitoring within domestic environments. This study presents a comprehensive comparative evaluation of three deep learning paradigms Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNN), and hybrid CNN-LSTM architectures for the classification of behavioral anomalies using multi-modal sensor data. Leveraging the publicly available Single Elder Home Monitoring: Gas and Position dataset (UCI Repository), comprising 444,631 temporally-resolved instances (20-second intervals) of environmental gas readings (temperature, humidity, CO₂, CO, four MOX sensors) and binary positional infrared motion data across a residential layout, we develop and benchmark architectures capable of distinguishing normative activity patterns from potential safety-critical deviations. Our methodology incorporates rigorous preprocessing to mitigate sensor drift through Principal Component Analysis-based environmental correction, followed by architecture-specific feature extraction strategies. Experimental results demonstrate that the hybrid CNN-LSTM model achieves superior classification performance (F1-score: 0.942 ± 0.018) compared to standalone LSTM (0.911 ± 0.023) and CNN (0.887 ± 0.031) configurations, attributable to its dual capacity for spatial feature localization within the sensor array topology and temporal dependency modeling across sequential observations. The proposed framework operates exclusively on non-invasive, low-cost sensor modalities, thereby addressing critical privacy concerns inherent to video-based monitoring systems while maintaining robust detection sensitivity. These findings substantiate the viability of chemical sensing arrays coupled with multi-modal deep learning as a scalable, ethically-aligned foundation for next-generation aging-in-place support technologies.