xG-Regression-Model

Introduction

This project explores Expected Goals (xG) in football analytics using logistic regression. By engineering meaningful features and leveraging SHAP analysis, we developed a model that provides actionable insights into goal-scoring probabilities.

---

Dataset

The dataset includes detailed information on individual shots during matches:

• Spatial Features: Coordinates, distances, angles.
• Contextual Features: Headers, penalty area, defensive pressure.
• Dynamic Features: Nearest opponent distances, average distances of teammates/opponents.

---

Features

Some of the key engineered metrics:

• Distance to Goal: Measures the shooter's distance to the goal center.
• Pressure: Quantifies defensive pressure using distance-based decay.
• Header Angle Distance Metric: Combines angles and distances for header scenarios.
• Goal Distance vs. Opponent Distance: Evaluates the relative positioning of the shooter and closest defender.

---

Model Evolution

The xG model was developed iteratively:

1. Baseline Model: Basic features like distance_to_goal and angle_to_goal.
2. Intermediate Model: Added defensive metrics (pressure, nearest_opponent_distance).
3. Final Model: Integrated advanced metrics like header_angle_distance_metric and goal_distance_vs_opponent_distance.

---

Results

Calibration Curve

The final model demonstrates significantly improved calibration.

ROC Curve

The ROC curve shows an AUC of 0.80, outperforming baseline models.

SHAP Analysis

Key features influencing the xG model predictions:

---