Evaluation of Statistical and Machine Learning Approaches to Modeling Medical Time Series Data
This project focuses specifically on the problem of incorporating time-series clinical care data into clinical prediction and decision support tools. The objective is to to develop approaches that can interpolate extrapolate and reduce the dimensionality of the time series data for several different types of repeatedly measured clinical data. Methods with existing software will be used to model the time series data. Different approaches will be evaluated by cross validation and Monte Carlo simulation. We hypothesize that the optimal method will differ and depend on both the objective of the analysis and the clinical variable under consideration. Hence special care must be taken with sparsely measured data or when the frequency of clinical tests vary significantly between patients.

I took a SuperUROP because I wanted an intense research project requiring continuous commitment. This project is a good match for me because it builds both on the machine learning techniques I’ve learned as well as the statistical experience I’ve gained from past research projects in working with clinical data. Mostly I am excited by the opportunity to make an impact at the intersection of medicine and technology.