Statistical Methods and Computation for Sports Analytics

点击翻译按钮获取中文摘要

With the growing fascination in sport and the increased availability of data in both size and complexity, the opportunity to do exciting work in sports analytics has never been greater. One of the most widespread and important problems in sports analytics is player evaluation. In professional sport, player evaluation has an impact on awards, roster selection and the management of salaries. Player evaluation is one of the primary objectives in this research proposal. Player evaluation in many sports has traditionally relied on a myriad of statistics, not all of which provide an accurate picture of a player's contribution to the sport. What has changed today is the availability of massive amounts of data including player tracking data. For example, in some leagues of professional soccer, the location of the ball and the 22 players on the field are recorded 10 times per second over 90 minute matches plus added time. These massive datasets allow researchers to investigate nuances of the sport and paint a more accurate picture of the entirety of player contribution. Two sports that I plan on studying are the growing (but still minor) sport of pickleball and the most popular sport in the world, soccer. In pickleball, we have a large dataset from doubles events where the partners regularly change. Using match scores as data, we plan on providing accurate rankings of player strength where match outcomes are confounded by the individual contributions of players. Specifically, we are interested in the degree to which pickleball is a strong (weak) link sport; a strong (weak) link sport is one where the stronger (weaker) partner has a greater influence on the match outcome. In the sport of soccer, we will develop player evaluation methods that take advantage of detailed player tracking data and soccer insight involving activity "off the ball". It is tempting and convenient in sport to analyze event data - that is, what has happened in the vicinity of the ball. The consideration of what did not happen, characterizing why it did not happen and attributing worth in these circumstances is my intention. The idea is to replicate and enhance what a coach/manager infers when the coach closely follows the activities of a particular player. One of the technical aspects in the analysis involves the determination of similar formations via clustering methods so that worth can be assigned to player actions based on the outcomes from many similar formations. With our connections to national and provincial Canadian sports organizations, the research activities are important in providing a competitive advantage to Canadian teams. The methods are also novel and translate to other domains in the natural sciences and engineering. For example, our proposed clustering methods involve a novel temporal aspect in the sense that team formations are dynamic.