Zhihua Su Email & Phone Number

Gainesville, Fl, Us

Gainesville, Florida Area

Selected research projectsUse machine learning techniques to improve prediction in regression and classification • Derive a model with fused lasso regularization to handle regression and classification with both matrix and vector-valued predictors• Applied the model to predict the bike rental demand in Washington D.C.’s bike share system and reduced the prediction error by around 30% compared to existing models• Applied the model to classify if a patient is diabetic from dietary preferences over the years and demographics variables, and reduced misclassification error by 50% compared to existing models• This work is published in Electronic Journal of StatisticsSoftware development with R • Develop a fast computing algorithm for a class of new statistical models called envelope models • Implement the algorithm to an R package called Renvlp and publish it on CRAN such that researchers in data science or applied sciences can use the envelope models for their own applications• The package has over 23K downloads up till today• The fast computing algorithm is published in Journal of Multivariate AnalysisCollaborate with applied scientists and contribute in experimental design and data analysis stages • Collaborate with researchers in soil and water sciences and horticultural sciences in experimental design step to study the effect of phosphorus fertilizations on the yield of chipping potato in Florida.• Once the data is collected, analyze the data using model corresponding to the design and write the data analysis report• Research results is published on Agronomy, which provide guidance on level of phosphorus fertilization to Florida potato farmers.

Selected research projectsDevelop Bayesian envelope model • Construct a new class of prior distributions to incorporate prior information in the context of envelope models, which is a new class of statistical models for efficient estimation and better prediction• Establish the Bayesian envelope model and derive an MCMC algorithm to obtain its estimator• The model extends the applicability of envelope model to small sample size scenarios• This research is published in Annals of Statistics

Minneapolis, Minnesota, United States

Train new teaching assistants on teaching skills

Doctor Of Philosophy - Phd, Statistics

Bachelor Of Science - Bs, Mathematics