Markov model methods for discrete sequences and soccer analysis

Cardenas-Torres, Eduardo

Permalink

https://hdl.handle.net/2142/129912 Copy

Description

Title

Markov model methods for discrete sequences and soccer analysis

Author(s)

Cardenas-Torres, Eduardo

Issue Date

2025-07-01

Director of Research (if dissertation) or Advisor (if thesis)

Zhao, Dave

Doctoral Committee Chair(s)

Zhu, Ruoqing

Committee Member(s)

• Foss, Alexander
• Eck, Daniel J

Department of Study

Statistics

Discipline

Statistics

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Sports Analytics
• Classification
• Bayesian Context Trees
• Bayes Classification
• Soccer
• Parking the Bus
• Markov Models

Abstract

This dissertation develops new Markov model-based methods for analyzing sequences of discrete data, with applications to viral genomic sequencing data and for soccer data. The first method is to address how a specific defensive tactic can affect soccer games throughout various professional leagues around the world. The method for the viral genomic compression is a new Bayesian classifier that’s based on compression analytics that utilizes a new approach for classifying documents. Chapter 2 introduces the novel study of a defensive style of tactic in professional soccer called Parking the Bus by using Hidden Markov Models. This study examines the effectiveness of Parking the Bus in preventing goals and identifies the specific moments during a game when this strategy is most beneficial. Using Hidden Markov Models, we uncover latent defensive states and develop metrics from football event data to identify instances of Parking the Bus. Our findings indicate that while Parking the Bus can reduce the likelihood of conceding goals, the effectiveness of playing defensively varies significantly across countries. When implemented strategically, it may be associated with success in international soccer leagues. Chapter 3 introduces a novel approach of classifying documents by utilizing Bayesian Context Trees to label viral genomic sequences. In this work, we were motivated by ideas from compression to develop a new document classifier. One barrier is that some popular compression algorithms, such as the Prediction by Partial Matching (PPM) method, lack a rigorous probabilistic model and cannot be directly used for statistical classification. To resolve this issue, we will build an optimal Bayes document classifier using Bayesian Context Trees (BCT), which, similar to PPM, can incorporate variable context lengths. Our approach can model the dependencies between document context and class label effectively and illustrates the rich connections between compression analytics and classification.

Graduation Semester

2025-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129912

Copyright and License Information

Copyright 2025 Edcuardo Cardenas-Torres

Owning Collections