Create Your First Project
Start adding your projects to your portfolio. Click on "Manage Projects" to get started
ATP Sports Strategic Analysis
Location
Los Angeles, CA
Date
April 2026
Role
Unsupervised Learning - Factor Analysis & Implications
Project type
Sports Analytics Lab
Key Features & Methodologies:
Data Pipeline: Processed a decade of ATP match data, converting raw match statistics into single-player observations with advanced feature engineering (e.g., Aggression Index, Consistency Index, Pressure Differential).
Supervised Learning: Built and evaluated Decision Tree and Random Forest models (with 5-fold cross-validation) to predict semi-final appearances, identifying key statistical thresholds (e.g., first-serve win percentages) for tournament success.
Unsupervised Learning: Applied Principal Component Analysis (PCA) and Factor Analysis to discover surface-specific player archetypes (e.g., "Servebot," "Elite Clay Grinder," "Defensive Baseliner").
Strategic Insights: Translated model weights into actionable coaching strategies, such as focusing on defensive consistency and break rates for the French Open versus aggressive hold percentages for Wimbledon.
Written Report
This project leverages historical ATP match data (2016–2025) to uncover the hidden variables that drive deep tournament runs in men's professional tennis. The analysis is divided into two primary machine learning pipelines: a supervised learning approach to predict which players will advance to the semi-finals of a Grand Slam, and an unsupervised approach (PCA and Factor Analysis) to define the underlying skill profiles required for different court surfaces (Hard, Clay, Grass).
To account for extreme target variable imbalance—as only 3-4% of a 128-player draw reach the semi-finals—we utilized SMOTE and evaluated models based on F1-Score rather than raw accuracy. The Random Forest model proved highly capable of capturing complex style interactions, successfully predicting semi-finalists entirely blind to player rankings or past performances.
