Spor analitiği için veri güdümlü performans analiz çerçevesi

Performance evaluation is a challenging, multidimensional and multi-criteria assessment problem. One application area is the player transfers in football (soccer), where player performance must be evaluated in-line with their responsibilities on the field. In this area of study, raw player performance statistics are not representative because of the external factors impacting the performance such as time-played, injuries, competition difficulty and characteristics, strength of the opponent, impact of actions in the game as well as the positions played. In addition, transfer market has unique financial dynamics in terms of transfer fees and player valuation. Some of the factors that affect transfer fees are athletic performance, properties of clubs and competitions and player popularity. The rich set of factors makes modelling transfer fees a challenging machine learning problem. This thesis provides a dynamic, context-dependent, probabilistic and hierarchical bottom-up approach for evaluating performance under uncertainty for custom requirements. Furthermore, the proposed framework links the performance metrics and various data sources to model transfer fees using machine learning ensembling methods. The proposed framework is generic and it can be adapted to other team sports.