Stephen has just completed the penultimate year of a Bachelor of Science (Statistics) degree at RMIT University in Melbourne, VIC. In 2014 he has completed work placements with both the Port Melbourne Football Club collecting game day statistics, and with the Australian Football League working on a project to monitor the game day activity of runners.
Modelling NHL Data – Prediction, Performance and Analysis of Ice Hockey’s Big Data
Background. Sports Analytics involves the analysis of large data sets, with the volume of games and mountains of recorded information, analysing sport has never been so detailed. Recently, a project team has formed to bring together all of this data for NHL (North America’s premier ice hockey league), and utilising data from over ten seasons the challenge is now how to handle such large data in a meaningful manner.This project is about building models for analysis and prediction of performance of America’s premier winter sport. Working with the research team, Stephen will build baseline, team specific models in order to determine the appropriate approach to investigate such data. A major challenge is the determination of what variables to include, as the data is expansive in its number of variables and its size.Stephen shall gain experience in using baseline sports modelling process (Elo, exponential smoothing, Tukey smoothing etc.) to determine simple predictive models. Then he shall utilise a variety of approaches to discriminate variables both statistically, and qualitatively. He shall explore advanced sports model, such as hybrid models and potentially conditional distribution modelling. He shall conclude on the approaches usefulness against key performance measures such as MSE, strike rates and return on investment.
Research Questions. What are the variables requiring inclusion in the modelling process? How do we decide which to include and what methods do we use to validate these decisions? What is the baseline model and how does it perform? What variables improve the key targets, such as strike rate, error margin and return on investment? How do we predict game outcome utilising score outcomes? How do model in-play characteristics?