Scalable Healthcare Analytics in the Cloud: Applying Bayesian Networks, Genetic Algorithms, and LightGBM for Pediatric Readmission Forecasting

Abstract

Pediatric readmissions present significant challenges in healthcare, leading to increased costs, poor
health outcomes, and psychological stress on families. Traditional prediction models often struggle
to handle the complexity and variability of healthcare data, which limits their predictive
capabilities. In response, this study explores the potential of integrating AI and cloud computing
to enhance prediction accuracy and scalability for pediatric readmissions. The objective is to
compare three machine learning techniques—Bayesian Networks, Genetic Algorithms, and
LightGBM—in a cloud-based environment to determine the most effective method for predicting
pediatric readmissions at scale. Bayesian Networks are used to model complex interdependencies
between patient characteristics, while Genetic Algorithms are applied for feature optimization and
selection. LightGBM, known for its speed and accuracy in handling large datasets, provides fast
predictions with high accuracy. The models are implemented on a cloud platform, enabling
scalable, real-time predictions using diverse data sources, such as electronic health records and
patient demographics. The results show that the AI-powered cloud computing model outperforms
individual methods, achieving an accuracy of 88%, precision of 85%, recall of 83%, F1-score of
84%, and an AUC of 92%. This combination of machine learning techniques significantly
improves prediction accuracy, scalability, and real-time performance in healthcare applications.
The study concludes that integrating Bayesian Networks, Genetic Algorithms, and LightGBM
within a cloud environment offers a robust solution for predicting pediatric readmissions, making
it a valuable tool for reducing unnecessary hospitalizations, optimizing care delivery, and
improving patient outcomes in pediatric healthcare systems.