Predicting Complications of Cirrhosis using Synthetic Data Generation Enhanced Dynamic Classifier Selection

Early identification of the risk of complications in cirrhosis is essential for facilitating timely interventions, personalized treatment plans and better patient management. However, developing data-driven models using real-world Electronic Health Records (EHRs) data is often hindered by challenges related to data accessibility, integrity, sparsity and class imbalance. Synthetic data that captures the characteristics and patterns of real-world EHRs data can be used for model development and evaluation that effectively addresses these data-related issues. In this study, we propose a dynamic classifier selection-based framework that incorporates both real-world EHRs data and synthetic data generated by generative adversarial networks. The framework trains a diverse set of heterogeneous classifiers, leveraging the benefits of both data sources. By utilizing dynamic classifier selection, the framework adaptively selects the best classifier based on the specific characteristics of the input data. The proposed framework was assessed through experiments to evaluate its effectiveness in the early prediction of the prevalence of acute on chronic liver failure and hepatic encephalopathy. The experimental results consistently demonstrate the superior performance of the proposed framework compared to baseline approaches. This study showcases the potential of the framework to improve clinical predictions and contribute to advancing our understanding of disease progression.