Using baseball pitch predictions, Hickey and Tao explored machine learning models and how to improve interpretability of results.
Fairfield Dolan Master of Science in Business Analytics (MSBA) alumnus Kevin Hickey has accomplished what few students at the graduate level achieve. Earlier this year Hickey, in collaboration with Jie Tao, PhD, assistant professor of information systems and operations management, won the 2020 Hawaii International Conference on System Sciences (HICSS) Best Paper Award in the Collaboration Systems and Technology track.
Hickey’s paper, “Dissecting Moneyball: Improving Classification Model Interpretability in Baseball Pitch Prediction,” was selected from over 100 papers, many of which were contributed from individuals with their PhDs. Hickey and Tao collaborated with Lina Zhou, PhD, professor of information systems and operations management in the Belk College of Business at UNC Charlotte.
The award was presented by a top researcher in the field, Jay F. Nunamaker, PhD, of University of Arizona. According to Dr. Tao, Dr. Nunamaker expressed that he was impressed by Hickey’s work and the fact that a graduate student was “so keen and dedicated to such a cutting-edge topic in the field.”
HICSS is the longest standing scientific conference in the field of information systems and technology and recognizes innovation and advancement at the global scale. Hickey’s paper was the result of his MSBA capstone course project which originated in the work he did as part of Dr. Tao’s data mining course.
“In machine learning there is a trade-off between interpretability of models and developing complex black boxes,” explained Dr. Tao. “We try and decode those black boxes, and in Kevin’s work, he picked baseball as a case study to illustrate our methodology.”
“Some of my first memories are of baseball,” said Hickey. He started playing Little League Baseball, continued through high school, and then attended University of South Carolina Upstate where he played Division 1 Baseball as an undergraduate. In his data mining course he learned to build machine learning systems through models to classify certain tasks, whatever the domain; he thought it would be interesting to apply the methodology to baseball.
Hickey and Tao explored two methodologies: LIME, local interpretable model-agnostic explanations, and SHAP, shapley additive explanations. In their paper they combined LIME and SHAP techniques together with pitching data to predict the resulting location of a pitch thrown by a pitcher and how this information might be used to train pitchers. What makes their findings all the more exciting is that their methodology has great potential to be used in fields other than sports, for example, in healthcare or finance. At present, Hickey and Tao are working on the extended version of the study to triangulate the analysis and Hickey is pursuing a PhD program.
“Receiving a Best Paper Award as a master’s student, at such a prestigious conference, can certainly open doors,” said Dr. Tao.