Giyilebilir sağlık sensörleri için makine öğrenimi destekli replikasyonsuz depolama optimizasyonu

Özet (abstract)

Wearable sensors (WS) are becoming an important tool for detecting and monitoring physical changes in humans, which, together with improved diagnostic techniques and clinical procedures, represents an important development in the clinical field. The information obtained from these sensors is dynamic and dependent on human activities, making efficient storage crucial for subsequent analysis. However, if data quality management is not properly handled, strategic process failures can reach up to 40%. At the same time, scalability concerns can make it difficult to scale data processing without affecting consistency and security. Moreover, replicated entries in wearable sensors can be indexed at multiple points, increasing storage requirements. Another widespread problem that may affect the accuracy and reliability of data management systems is the absence of unique records. To address these problems, this work introduces Replication-Free Data Management (RDM) to ensure reliable data storage in healthcare systems. The proposed method is based on the Random Forest classification algorithm, which aims to differentiate between aggregation and classification instances. This approach helps minimize data loss caused by extended latency by categorizing activities and verifying their similarity. Conditional scheduling based on activity instances and scheduling slots is also used to distinguish similar sensor data from non-comparable data during storage access. Experimental results demonstrate that RDM supports different scheduling times while maintaining 0.0816 redundant data and 419.61 ms latency and supports different scheduling instances while maintaining 0.0831 redundant data and 426.58 ms latency. Furthermore, RDM achieves 92.28% storage utilization and 6.23% data loss for different classification instances, and 92.44% storage utilization and 6.28% data loss for different aggregation times. The proposed RDM achieves 97.12% performance and 98.43% efficiency ratios, demonstrating its effectiveness in reducing data replication, data loss, and latency across various scheduling intervals and instances.

Yazarların özeti; kaynağından alınmıştır. Journal Of Big Data, 2026 · DOI ↗