FGOE Pro: Decoupling NFL Kicker Skill from Situational Variance using Amortized Variational Inference
A Bayesian state-space modeling framework that evaluates NFL kicker performance by calculating Field Goal Over Expected (FGOE) to decouple environmental difficulty from skill degradation.
Machine Learning
Python
Project Overview
Project Type: Sports analytics framework and probabilistic evaluation model.
Technologies Used: Python, PyMC / Stan, Bayesian Statistics, State-Space Models, Data Analytics.
Timeline: August 2026 - Present
My Role: Lead Data Scientist & Researcher.
Overview: Developed a Dual-Process Bayesian State-Space Model designed to rigorously evaluate kicker performance in high-pressure football scenarios. The model processes fine-grained situational context—including kick distance, clock pressure, score differential, and career volume—to compute a normalized Field Goal Over Expected (FGOE) metric. By isolating situational difficulty from athlete variance, the system accurately distinguishes between high-risk misses driven by environmental factors and true latent skill degradation over time.
How to Access:
Live Demo / Interactive App: https://kicker-analysis-dashboard.vercel.app/
GitHub: [Currently Private]