Streaming de dados de sensores vestíveis: a construção de um data warehouse near real time

Abstract

Population aging and the need for continuous clinical monitoring demand advanced technological solutions in digital health (e-health). This work presents the implementation of a Near Real-Time (NRT) Data Warehouse designed for processing data streams from wearable sensors. The proposed architecture utilizes Apache Kafka for data ingestion, Apache Flink for stream processing with 10-second time windowing, TimescaleDB for optimized time-series storage, and Grafana for real-time analytical visualization. The solution was validated using the HAR70+ dataset, which comprises inertial accelerometry data from elderly individuals monitored during activities of daily living, such as walking, sitting, and lying down, over a two-month period. The results demonstrated a processing latency between 10 and 15 seconds from signal capture to visual availability, representing an approximately 99.9% reduction in informational latency compared to traditional batch ETL models (24-hour cycles). The implemented dimensional modeling enabled efficient analytical queries regarding the users' motor performance, while the Lambda Architecture balanced low-latency processing with reliable high-volume historical persistence. The system enables remote and continuous monitoring of the functional capacity of the elderly, allowing for rapid interventions and providing a robust infrastructure for future predictive systems based on machine learning.