Research
Decision-Making under Uncertainty with Batch Data
The increasing availability of data brings new opportunities to improve decisions. In applications with high-stake decisions, we are often not allowed to perform experiments, but given only a batch of data and have to find a good policy. Examples include healthcare, education, autonomous driving, and hazard management, a concrete example of which is wildfire management. I develop new methods and establish theoretical results for such decision-making problems.
Data Analytics for Wildfire Operations
Wildfires can have devastating impacts on our society. Due to an increase in global temperature and extreme weather, the frequency and severity of wildfires have increased in recent years and are forecasted to continue to increase. I apply data analytics to help wildfire managers make better decisions in operating suppression resources, such as helicopters and firefighters.
Data Analytics for Healthcare
I apply data analytics to help clinicians, hospital managers, and policy makers improve patient outcomes and to make care delivery more efficient.
For example, I recently worked on readmission prediction. Among patients hospitalized in Canada, approximately 9% are readmitted within 30 days. Reducing readmissions can improve not only the patients’ health outcomes but also the efficiency of hospitals and the overall health system. However, most existing prediction models fail to predict readmission well. My research team built prediction models using extensive data of more than 468k patients over seven years. We use deep learning methods and skip-gram to construct features automatically from the longitudinal data. Our findings show that these features improve the performance of various prediction models substantially.