Cloud-Assisted Machine Learning Approaches for Predicting Diabetes Progression from Real-World Data

Abstract

This study introduces a cloud-enabled machine learning framework to predict diabetes progression via synthetic patient data. Utilizing sophisticated simulation methods to create a large set of simulated patient data to mimic all aspects of actual clinical data (e.g., HbA1c levels, BMI, age and length of time since diabetes diagnosis), this research has tested several machine learning techniques, such as Random Forest, XGBoost and Long Short-Term Memory (LSTM) to develop and test predictive models. Additionally, a cloud environment allowed for efficient, on-demand prediction, and frequent retraining of the model in order to ensure that the model is updated based on new input data. Results from the evaluation of the predictive capabilities of the developed models indicated excellent predictive performance and reliability; this suggests that a cloud-based system may be beneficial to diabetes care through providing clinicians with timely and data-driven information for decision-making.

Ähnliche Arbeiten