Football Prediction Model

^{Developed a model aimed at predicting NFL game outcomes. Employed advanced statistical techniques to analyze vast datasets.}

Skills

Replit
Pycharm
Data collection via REST APIs

Tools

Languages: Python/JS
Libraries: Pandas, Scikit-learn, NumPy, Matplotlib

Project Overview

A Point-Spread Equivalent Power Rating System. Quantitatively evaluating every element that could impact a game's outcome — from individual players and coaches to division rivalries — and assigns them point-spread equivalent values. These values are then summed to generate a power rating for each team, which in turn informs our game predictions.

Data Collection & Analysis

Collected historical NFL data (via ESPN and X.com API)
Applied Python-based statistical methods to identify key performance indicators affecting game outcomes. Organized data and rule sets using Pandas for easy retrieval and further analysis.

Rule Definition & Weighting

Developed a weighted scoring system to assign different levels of importance to key performance indicators like Points per Game, Field Goal Percentage, and Turnovers.
Created a set of custom heuristic rules influenced by these weighted scores, leveraging them to make more precise predictions.
Utilized Python scripts to run simulations on historical data, optimizing both the rules and their corresponding weights for better prediction accuracy.

Evaluating & Iterating

Prediction vs Reality: After each game, I'd compare our model's predictions with the actual outcomes. If the predictions are consistently off, I'd recalibrate the heuristic rules and weighting systems.
Rule Adjustments: Dynamic updating of our point-spread equivalent values and power ratings based on the season’s unfolding events—trades, injuries, and player performance trends.
Game Week Reviews: Weekly dives to study the performance of our model, taking into account upsets, blowouts, and other anomalies to fine-tune.

Bankroll Management & The Kelly Criterion

Optimal Sizing: The Kelly Criterion is a mathematical formula that calculates the optimal bet size based on your predicted winning probability and the odds offered.
Automated Calculation: Our model integrates the Kelly Criterion to auto-calculate optimal bet sizes based on the odds that the user inputs into the console + our prediction model's outputs.
Risk Mitigation: Using the Kelly Criterion, we mitigate our risks by aligning bet sizes with both the edge and the bankroll, thereby maximizing long-term gains while minimizing the risk of busting our bankroll.