| dc.description.abstract | This thesis describes analysis done to investigate predictions on injury time in a football game. Statistical and several machine learning techniques have been applied to predict how many minutes will be added by the referee at the end of each half. This research has been done in cooperation with a company called Smartodds, who provides statistical research and sport modeling services for a betting syndicate. The thesis consists of a literature review of football modeling, description of the methods applied, handling and assessment of dataset, provided by Smartodds, results and comparison of the models, and a discussion of the results and a conclusion. Four different models have been developed, a linear model, a Poisson model, a negative binomial model and an artificial neural network model.The performances of the models are compared, and there is not much separating one from another, in the end all of the models are rejected by a χ2 goodness of fit test. By a variety of reasons it might be impossible to achieve accurate point predictions of injury time.This can be caused by the incompleteness in the dataset or simple a non-recurrent behavior of the data making it impossible to predict with sufficient confidence based upon neither statistical methods nor machine learning techniques | |
