Enhanced automatic feature extraction in heterogeneous biomedical time series for optimization of clinical monitoring and diagnosis

RWTH Publications (RWTH Aachen)

Abstract

The prevalence of real-time data acquisition has increased significantly in recent years due to advancements in sensors, transmission, and storage technologies. This has led to the collection of large datasets in various fields, including engineering, medical, physical, and social sciences. In healthcare, dynamic monitoring data has become a crucial aspect for health diagnosis, treatment, and proactive health tracking. The emergence of wearable technology and monitoring systems has provided new avenues for understanding complex dynamic processes through longitudinal data, providing insight into an individual's health and comfort. However, it is important to note that data quality issues can significantly impact temporal data analysis. Additionally, the analysis strategies required for monitoring data vary greatly between monitoring a few parameters at high sampling rates and monitoring several parameters over extended periods at low sampling frequencies. One approach for improving such data analysis is to represent the data as features, enhancing the interoperability of outcomes from statistical and machine learning analyses. This thesis explores the domain knowledge feature engineering of monitoring time series data through a Python package, developed to provide a robust and systematic temporal data analysis platform. The package includes functional elements ranging from data reading to denoising and feature extraction, and its modular design allows for diverse applications and the incorporation of user-defined components. The findings of this thesis demonstrate the capability of temporal data abstraction using a methodical feature extraction scheme. The developed workflow provides a robust and user-friendly architecture for conducting time series analysis.

Authors 1

1. Pejman Farhadi Ghalati corresponding Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen