Advanced Machine Learning for Data-driven Disease Prediction

The rapid advancement in sensing and information technology has ushered us into an era of data explosion, where a large amount of data is now easily available and accessible in the clinical environment. This wealth of healthcare data offers new avenues for developing automated data-driven methods for disease diagnosis. Electronic Health Records (EHRs), serving as digital repositories of a patient's medical information, present unique opportunities to analyze and decipher clinical events and patterns within large populations. Given the rich information about a patient's health trajectory, leveraging EHRs through data-driven methodologies can significantly enhance clinical decision support systems.

However, utilizing real-world EHRs for reliable data-driven disease detection presents several challenges due to the observational nature of EHRs. Unlike well-defined, randomized longitudinal experiments, EHRs are recorded only when patients receive care, resulting in complex and highly heterogeneous data. This dissertation focuses on developing robust and reliable advanced machine-learning algorithms to investigate complex EHR data for disease prediction, addressing the challenges of diverse and complex data types, incomplete datasets, and imbalanced class distributions in EHRs.