[Pycon] [new paper] "Michał Górnik" - Beating the bookies in football – predicting matches with Machine Learning

[info a pycon.it]

Title: Beating the bookies in football – predicting matches with Machine Learning
Duration: 45 (includes Q&A)
Q&A Session: 15
Language: en
Type: Talk

Abstract: Huge amount of money circulating in sports betting market makes predicting results in a sport competitions attractive research problem. During the talk we are going to briefly describe both: classification and regression approaches for predicting football matches outcome as well as features that can leverage model accuracy. We will describe how to design and implement typical Machine Learning flow on the example of football matches prediction with Python – process of incoming data acquisition, feature extraction, model retraining and providing betting strategy.

We are going to present various types of data that can leverage algorithms accuracy – historical team performance, players market value and skills or even matchday weather – with the use of tools for ML models interpretation we will assess how strong impact particular features have on chances that particular team wins.

After the feature engineering process, we will speak about ways of selecting the best among multiple models - due to sequential nature of football data in time, based on time hold-out cross-validation will be described for choosing best model structure and hyperparameters. We are going to show how to design a betting strategy based on the best classification model and compare it against odds data from most popular bookmakers.

At the end we will describe how to design and implement typical Machine Learning flow on the example of football matches prediction with Python – process of incoming data acquisition, feature extraction, model retraining and providing betting options.

The talk will be presented by both of us: Michał Górnik and Jacek Krasnoborski



Tags: [u'classification', u'machine-learning', u'scikit-learn', u'regression', u'sports-analytics']