Markov Decision Process Approach to Stability-Aware Decision Tree Modeling for Temporal Healthcare Prediction

This project investigates the integration of Markov Decision Processes (MDP) with Decision Tree-based learning to enhance model stability and interpretability in clinical decision-making. It focuses on balancing predictive performance and structural robustness, addressing one of the key limitations of traditional decision trees. Additionally, the research explores techniques to embed patient-level temporal dynamics within the model structure, enabling early and explainable prediction of critical conditions such as sepsis using publicly available ICU datasets.

Towards Human-Centered Decision Tree Models for Reliable Clinical Predictions

This project explores a novel decision tree generation approach that integrates expert knowledge and human-like reasoning to improve stability and interpretability in clinical decision-making. By guiding the tree construction process through expert-informed feature prioritization, the method aims to mimic the way clinicians evaluate key factors first, leading to more efficient, accurate, and explainable predictive models for healthcare applications. For instance, in a medical dataset for detecting fever, the model prioritizes key features such as body temperature first, similar to how a clinician would approach the problem.