For his day job at Battelle, statistician Andrew Landgraf works on a variety of data science problems ranging from clinical predictive analytics to brain signal interpretation. In his spare time, he uses his analytical expertise to tackle data science contests across many different domains.
Recently, Andrew placed second out of 73 teams in the defined-data qualifying match of the Global Energy Forecasting Competition (GEFCom) 2017. The competition, which was sponsored by the IEEE Power and Energy Society, asked competitors to predict energy use for 10 New England regions on an hour-by-hour basis across a four-month period from January 2017 to April 2017. His solution ensembled statistical techniques including quantile regression, gradient boosting machines and random forests to produce a probabilistic prediction for each hour within the time frame.
As a result of his winning prediction, Andrew was invited to present his methods at the International Symposium on Energy Analytics in Cairns, Australia, and his methods will be published by the International Institute of Forecasters in a special journal issue. Building better medium-range predictions for energy use will help energy planners make better decisions that will improve the stability of the electrical grid.
This is not Andrew’s first data science competition win. In April of this year, he won a competition to predict the outcomes of the college basketball championship, and in 2013 he won a student competition hosted by Capital One to give their customers the most accurate personalized recommendations for new places to shop.
He explains, “These are all machine learning problems. The data science and statistical methods are very similar, no matter what industry you are working in. We just need to apply domain expertise to understand the objective and define the best inputs for the model.”
Andrew has applied his expertise to a broad range of problems at Battelle, including statistical analysis for the federal highway administration and chemical and biological warfare agent identification. In the health analytics realm, he worked on data analysis for the Healthy Communities Study, a five-year study of factors influencing childhood obesity. He was also part of the analytical team for Battelle NeuroLife™, a groundbreaking neural bypass technology that interprets brain signals from the motor cortex and sends them to a specialized sleeve to give paralyzed patients conscious control of their hands. For another project, he built a machine learning model with clinical and proteomic data to predict the likelihood of a patient needing a knee replacement and identified the most important contributing factors.
Andrew holds both a Ph.D. and an M.S. in Statistics from The Ohio State University along with a B.S. in Actuarial Science. His research has focused on machine learning using high-dimensional count and binary data. His past experience includes transportation research with the Campus Transit Lab at Ohio State University and forecasting electricity usage at IGS Energy. He also was a fellow in the Eric & Wendy Schmidt Data Science for Social Good Fellowship at the University of Chicago.
He looks forward to applying his expertise to new problems, both at Battelle and for future data science competitions.