book

Soccer Analytics with Machine Learning

by Haipeng Gao, Ari Joury, Weining Shen, Guanyu Hu

June 2026

Intermediate to advanced

344 pages

8h 14m

English

O'Reilly Media, Inc.

Read now

Unlock full access

Preface
Who Should Read This BookWhy We Wrote This BookNavigating This BookConventions Used in This BookUsing Code ExamplesO’Reilly Online LearningHow to Contact UsAcknowledgments
1. The Soccer Analytics Landscape
The Problems That Soccer Analytics Helps SolveSoccer Clubs: Making Decisions Across the OrganizationBetting Markets: Modeling Uncertainty and Pricing OutcomesFans and Analysts: Reading the Game Beyond the ScorelineThe Data-Driven Evolution in SoccerFrom Outcomes to Process: The Limits of Traditional AnalysisThe Data Revolution: From Event Logs to Tracking SystemsAnalytics in Practice: Metrics, Applications, and ScopeWhy Soccer Resists Easy AnalysisA Framework for Soccer AnalysisThe Three Types of AnalysisData Sources for Modern AnalysisFrom Player to League: Levels of AggregationPrinciples That Run Through EverythingLooking Ahead: A Guided Tour of the BookConclusion
2. Python Fundamentals: Building Your Analytics Toolkit
Python: The Language Behind Modern AnalyticsYour Analytics Workspace: Setting Up the EnvironmentInstallation: Getting Python ReadyVirtual Environments: Isolating Your ProjectsJupyter Notebook: Creating a Workspace for Interactive Analysis and MoreRepositories: Storing Your Project FilesPython Basics: The Core MovesPrimitive Data Types: Representing Basic InformationBasic Operations: Working with Values and ComparisonsVariable Assignment: Storing and Reusing InformationData Structures: Organizing InformationConditional Statements: Writing Decision RulesLoops: Repeating Computation EfficientlyList Comprehensions: Performing Compact and Pythonic IterationFunctions: Providing Reusable Logic for AnalysisLooking Ahead: From Python Basics to Real AnalysisConclusion
3. Exploratory Data Analysis in Soccer
Understanding Soccer Data LevelsMatch-Level DataPlayer-Level DataEvent-Level DataWhat Is EDA in Soccer Analytics?Loading StatsBomb Soccer DataGetting the DataLoading the DataTaking a First Look at the DataStructured EDA by Data TypesNumerical Features: The Quantifiable GameCategorical Features: The Qualitative AspectsOrdinal Features: The Ordered CategoriesPreliminary Exploration: Getting Your BearingsBuilding Tournament DataFramesCounting the Frequency of Event TypesFiltering: Passes by Player or TeamGrouping: Per-Match Pass SummariesAggregating: Shots, Goals, and xG by TeamBasic Data Cleaning for SoccerHandling Missing ValuesStandardizing NamesVisualizing Soccer DataA Quick Note on Matplotlib and SeabornBar Charts and RankingsHistograms and DistributionsBoxplots and VariabilityTime Series and Cumulative PlotsScatterplots: Shot Volume Versus GoalsSoccer-Specific Visualizations with mplsoccerPass Location HeatmapsAdvanced Pitch Visualizations: Passing NetworksBasic Pitch VisualizationsAdvanced Heatmap VisualizationsCase Study: Japan Women’s at the 2019 World CupTeam-Level OverviewMatch FlowSpatial PatternsPlayer-Level InsightsWrap-up of the Case StudyFrom EDA to Machine LearningFeatures as Building BlocksThe Standard PipelineLooking Ahead: From Exploration to RegressionConclusion
4. Regression Techniques for Soccer Analytics
From Lines on a Field to Lines of Best Fit: Linear RegressionPredicting Market Value: A Player’s WorthBuilding the Model: The Line of Best FitVisualizing the ModelEvaluating Performance: How Good Is Our Model?Counting Goals: Poisson RegressionPredicting Goals in a MatchVisualizing the Poisson ModelFinding Similar Players: K-Nearest Neighbors RegressionPredicting Player Value with KNNChoosing the Right kEvaluating and Interpreting Regression Models in Soccer ContextsUsing Key Performance MetricsUnderstanding the Importance of a Validation SetInterpreting CoefficientsVisualizing KNN: Decision BoundariesComparing ModelsAdvanced Topics in Poisson Regression: Overdispersion and Zero-InflationOverdispersion: When Variance Exceeds the MeanNegative Binomial RegressionZero Inflation: Too Many ZerosHurdle Models: An Alternative ApproachModel Diagnostics: Ensuring That Your Model Is SoundResidual Analysis: Looking for PatternsGoodness-of-Fit TestsInfluential Observations: Outliers that MatterModel Selection: Choosing the Best ModelPractical Case Study: Predicting Match OutcomesGenerating the DatasetPerforming Exploratory AnalysisBuilding and Comparing ModelsEvaluating on Test DataInterpreting the ResultsFeature Engineering for Better PredictionsInteraction FeaturesPolynomial FeaturesCategorical FeaturesLagged FeaturesDomain-Specific FeaturesCommon Pitfalls and Practical TipsPitfall 1: Overfitting to the Training DataPitfall 2: Ignoring MulticollinearityPitfall 3: Using the Wrong Model for the DataPitfall 4: Forgetting About Data QualityTip 1: Start Simple, Then Add ComplexityTip 2: Visualize, Visualize, VisualizeTip 3: Document Your ProcessTip 4: Validate with Domain ExpertsLooking Ahead: Where Regression Meets the Real WorldFrom Regression to ClassificationEnsemble Methods: Combining ModelsDeep Learning: The FrontierTime Series and Sequential DataCausal Inference: Beyond PredictionThe Human ElementConclusion
5. Predicting Soccer Outcomes with Classification
From Coin Flips to Goal Predictions: Understanding ClassificationLogistic Regression: Your First ClassifierThe Intuition: From Numbers to ProbabilitiesHands-On: Building an Expected Goals ModelMeasuring Success: How Good Are Your Predictions?The Confusion Matrix: A Report Card for Your ModelPrecision and Recall: The Two Sides of PerformanceThe ROC Curve: A Complete Performance PictureEvaluation of Probabilistic Models with Cost FunctionsBeyond Binary: Multiclass ClassificationK-Nearest Neighbors: Finding Similar SituationsThe Intuition: Birds of a FeatherHands-On: KNN for Shot PredictionChoosing the Right KCase Study: Building a Complete Match Outcome PredictorStep 1: Define the ProblemStep 2: Gather and Prepare the DataStep 3: Train and Evaluate the ModelStep 4: Interpret the ResultsKey Takeaways from the Case StudyPractical Tips and Common PitfallsHandling Imbalanced DataSelecting and Engineering FeaturesAvoiding OverfittingChoosing When to Use Which AlgorithmTuning the Decision Threshold: Finding the Sweet SpotUnderstanding the Threshold Trade-OffFinding the Optimal ThresholdDeciding When to Adjust the ThresholdComparing Models: Which Classifier Is Best?The Model Comparison FrameworkHands-On: Model Comparison for xGVisualization of Model PerformanceAdvanced Topics: Going DeeperScaling and Normalizing FeaturesHandling Missing DataCombining Classifiers with Ensemble MethodsReal-World ConsiderationsData Quality and AvailabilityModel Maintenance and UpdatesEthical ConsiderationsLooking Ahead: From Linear Boundaries to Decision TreesConclusion
6. Advanced Classification Methods
The Classification Tree: Soccer’s Decision FlowchartThe Intuition: Learning the If-Then Rules of SoccerHands-On: Building a Tree to Predict Pass OutcomeInterpreting the Soccer IntelligenceUnderstanding Pros and Cons: When to Use a Single TreeTuning the Tree: Preventing OverfittingRandom Forest: The Wisdom of the CrowdThe Intuition: Ensemble LearningHands-On: Building a Random Forest for to Predict Match OutcomeInterpreting Feature ImportanceTuning Random ForestsUsing Out-of-Bag ErrorUnderstanding When Random Forests StruggleXGBoost: Learning from MistakesThe Intuition: Sequential LearningHands-On: XGBoost for xG PredictionPerforming SHAP Analysis: Explaining the Opaque ModelTuning XGBoost: The Art of Hyperparameter OptimizationKnowing When to Use XGBoost Versus Random ForestPractical Applications: From Theory to PracticeExample 1: Building a Comprehensive xG ModelExample 2: Predicting Player ActionsExample 3: Recognizing Tactical PatternsHandling Imbalanced Data in SoccerUsing Feature Engineering for Tree-Based ModelsLooking Ahead: From Handcrafted Rules to Learned RepresentationsConclusion
7. Deep Learning for Soccer Analytics: From Intuition to Implementation
Why Deep Learning for Soccer? Beyond Traditional ModelsLooking Inside the Model: How Neural Networks Capture PatternsLearning from Data: Training, Loss, and OptimizationBuilding Your First Soccer Prediction Model with PyTorchBinary Outcomes: Will the Home Team Win?Match Results: Win, Draw, or LossGoal Prediction: Modeling Scoring OutputFull Match Simulation: From Goals to OutcomesA Second Lens: Implementing the Model in TensorFlowBinary Outcomes: Will the Home Team Win?Match Results: Win, Draw, or LossGoal Prediction: Modeling Scoring OutputFull Match Simulation: From Goals to OutcomesLooking Ahead: From Flexible Models to Smarter InputsConclusion
8. Feature Engineering for Soccer Analytics
Feature Engineering: From Raw Data to Predictive SignalsPerformance Features: The FundamentalsGoals For and Goals Against: Measuring Scoring PerformanceVenue Splits: Considering Home Versus AwayShot Quality: Evaluating Chance CreationTactical Features: Describing How Teams PlayPassing and Possession: Measuring Ball ControlTackles, Interceptions, and Recoveries: Tracking Defensive ActionsPressing Intensity: Applying Defensive PressureSet Pieces: Capturing Dead-Ball SituationsContextual Features: Incorporating Match ConditionsRest and Scheduling: Modeling Fatigue and CongestionForm and Momentum: Capturing Recent Performance TrendsRanking Features and Advanced Ranking Algorithms: Estimating Team StrengthThe Colley Matrix: A Linear Algebra ApproachPageRank: A Network-Based PerspectiveElo Ratings: A Dynamic Rating SystemA Comparison of Methods: Strengths and Trade-OffsPlayer-Level FeaturesFeature Engineering Pipeline: From Data to Model InputCase Study: Predicting the 2019 Women’s World Cup FinalLooking Ahead: From Better Inputs to Harder DecisionsConclusion
9. The Art and Science of Soccer Betting
Soccer Betting BasicsTypes of Soccer BetsOddsArbitrage OpportunitiesThe Real Dataset Used in This ChapterChoosing a SideAlways Pick the Home TeamAlways Pick the Lowest OddsUse a Simple Machine Learning BaselineBetting StrategiesFixed-Stake BettingKelly CriterionFrom Predictions to Potential ProfitThe Reality of ProfitabilityPractical ConsiderationsThe Importance of DisciplineThe Role of TechnologyEthical ConsiderationsLooking Ahead: Building on Top of the FundamentalsConclusion

10. What’s Next for Soccer Analytics?
What This Book Was Really AboutFrom Questions to DataData Preparation Is Part of the AnalysisExploration Before PredictionFeature Engineering Is Where Domain Knowledge EntersModels Support Judgment, They Do Not Replace ItThe Next Stage in Soccer AnalyticsFrom Static Datasets to Living Data PipelinesFrom Team-Level Models to Player-Level and Lineup-Level ModelsRicher Data: Tracking, Video, and Multimodal AnalysisStronger Modeling DirectionsFrom Notebooks to Analytical ProductsAI and Large Language Models in the Analytics WorkflowChallenges and LimitsData Quality and AccessInterpretability and TrustUncertainty Will Not DisappearEthics, Privacy, and Responsible UseBeyond SoccerA Practical Roadmap After This BookFinal Thoughts
Appendix. References and Further Reading
Further Reading by TopicFoundational Statistics and RegressionAdvanced Statistical MethodsMachine Learning and Deep LearningSports AnalyticsSoccer-Specific AnalyticsSports Betting and Business AnalyticsCited References
Glossary
Index
About the Authors

Content preview from Soccer Analytics with Machine Learning

Appendix. References and Further Reading

This appendix serves as a curated guide to the literature that underpins the methods and concepts discussed in this book. It is divided into two main parts: a collection of recommendations for further study, organized by topic, and a list of works cited directly in the text. Each entry in the further reading list includes a brief annotation to help you understand its relevance and what you can expect to gain from it.

Whether you are looking to deepen your understanding of a specific statistical technique, explore the broader field of machine learning, or dive into the nuances of soccer-specific analytics, this guide provides a roadmap for your continued learning journey.