Machine Learning for Preventive Healthcare in Seniors: Forecasting Hospitalizations Using Predictive AI : An Audio-Based Approach to Early Detection and Intervention

Sammanfattning

The increasing adoption of connected technologies and AI in healthcare offers new opportunities to improve senior care by enabling early detection of health risks and emergencies. This thesis explores the development of a non-invasive, sound-based AI system designed for passive health monitoring of elderly individuals living independently at home. Unlike wearable devices, the proposed system analyzes ambient audio to detect distress situations such as falls and cries for help, as well as to identify behavioral changes indicative of health deterioration. Using supervised deep learning and a hybrid edge- cloud architecture, the system achieves real-time detection while preserving privacy through on-device processing. Evaluation on real-world data confirms the feasibility of this approach, although challenges remain related to data quality, environmental noise, and model generalization. Future work includes expanding the dataset, improving audio quality, and enhancing temporal context analysis. This work contributes to safer aging-in-place solutions and demonstrates the potential of sound-based AI for proactive senior healthcare.