Uncertainty Management for Wearable IoT Wristband Sensors Using Laplacian-Based Matrix Completion

Contemporary sensing devices provide reliable mechanisms for continuous process monitoring, accommodating use cases related to mHealth and smart mobility, by generating real-time data streams of numerous physiological and vital parameters. Such data streams can be later utilized by machine learning algorithms and decision support systems to predict critical clinical states and motivate users to adopt behaviours that improve the quality of their life and the society as a whole. However, in many cases, even when deployed over highly sophisticated, cutting-edge network infrastructure and deployment paradigms, data may exhibit missing values and non-uniformities due to various reasons, including device malfunction, deliberate data reduction for efficient processing, or data loss due to sensing and communication failures. This work proposes a novel approach to deal with missing entries in heart rate measurements. Benefiting from the low-rank property of the generated data matrices and the proximity of neighbouring measurements, we provide a novel method that combines classical matrix completion approaches with weighted Laplacian interpolation offering high reconstruction accuracy at fast execution times. Extensive evaluation studies carried out with real measurements show that the proposed methods could be effectively deployed by modern wristband-cloud computing systems increasing the robustness, the reliability and the energy efficiency of these systems.