Title:  Early Sepsis Prediction for ICU Patients: A Reinforcement Learning Approach

Committee: 

Dr. Beyah, Advisor        

Dr. Xie, Chair

Dr. Kamaleswaran

Abstract: The objective of the proposed research is to build an online learning-based model for early sepsis prediction. There is plethora of early sepsis prediction methods which are based on machine learning. Since the dataset usually grows dynamically, most offline models using retrospective observational data cannot incorporate the new observational data to get better performance. Utilizing the new data to improve offline models requires retraining the model, which will result in a high computational cost. To solve this problem, an Online Artificial Intelligence Experts Competing Framework (OnAI-Comp) for early sepsis detection is proposed in this work. The existing experimental results demonstrated that OnAI-Comp converges to the optimal strategy in the long run. In addition, OnAI-Comp provides interpretable predictions using local interpretable model-agnostic explanation technologies.