Projects
Overview
Latent State Models for Personalized Medicine
-
Propose a partially observable Markov decision process framework that recommends personalized treatment for maximizing patient outcome.
-
Study asymptotics and causal inference in Q-Learning and V-learning with functional approximation.
Hidden Markov Models for Longitudinal Accelerometer Data
-
Develop a gradient-based estimation and inference for hierarchical hidden Markov model on zero-inflated count data.
-
Apply the methodology on clinical accelerometer data for arthritic cat and human physical activity.
Two Sigma: Using News to Predict Stock Movement
- Leverage news and market information to predict stock movement using neural network, random forest and gradient boosting tree. (Top 7% on Kaggle)
Data mining with the Truven Health Insurance Claim Database
-
Build a propensity-score matched logistic regression to identify risk factors associated with fracture nonunion in the highly unbalanced population of surgical and nonsurgical patients.
-
Implement multiple imputation methods for missing data.