Missing Sensor Data Handling for Wireless Body Area Network

Abstract

In healthcare, wireless body area sensor networks (WBANs) play a crucial role in monitoring patient activities for real-time health diagnostics. However, the reliability of these systems is challenged by the occurrence of missing sensor data, causing from various factors such as transmission issues, sensor malfunctions, physical conditions, and system level problems that are unavoidable. Addressing this challenge is important to guarantee the accuracy and usability of health monitoring applications. This thesis investigates strategies for handling missing sensor data within WBANs. Various mechanisms and patterns of missing data are explored, providing insights into the underlying causes and effects. A range of methodologies for handling missing data, including deletion methods, imputation methods, and approaches for retaining missing data, are discussed with a focus on imputation techniques. Experimental simulations are performed to observe the performance of different imputation models in estimating missing sensor data. The results indicate promising potential for imputing missing data using information from successfully observed sensors within the network. In the expriment, machine learning models such as K-Nearest Neighbour, Multivariate Imputation by Chained Equations, LightGBM Regressor, and CatBoost Regressor are implemented and compared to identify the suitable approach for handling missing sensor data in healthcare scenario. Overall, this thesis contributes to improve the understanding of missing data challenges in health monitoring systems and provides valuable insights into effective strategies for handling these challenges within wireless body area sensor networks.