Learning medical concept representations based on clinical data for hospital readmission prediction

This thesis aims to develop machine learning models to automatically predict patients’ hospital readmission. The proposed models can identify frequent patterns from patients’ hospital records to distinguish readmission and non-readmission cases. To address the sequential nature of patient hospital records, machine learning models that integrate patient clinical data,, demographic information, and external medical categorical data are proposed. Incorporating semantic information from medical data enhances readmission prediction performance. Additionally, the proposed models are capable of providing interpretable explanations that support the model’s decisions.