Abstract

This study applies Principal Component Analysis (PCA) combined with K-Means clustering to modern NFL quarterback performance data, incorporating the advanced metric Completion Percentage Over Expected (CPOE). Using 2025 season data and historical trends, we identified three robust archetypes: Dual-Threat, Balanced, and Pocket. Key findings show that Dual-Threat quarterbacks exhibit the highest fantasy points per game (mean ~22.8 FPG) with greater variance, while Pocket passers demonstrate the most consistent and predictable production (mean ~18.4 FPG). CPOE significantly improved archetype separation by highlighting precision differences beyond raw completion percentage. These data-driven archetypes provide actionable insights for fantasy football, team building, and player evaluation.

Introduction

The NFL quarterback position has evolved dramatically with the rise of mobile athletes and advanced analytics. Traditional classifications (pocket passer vs. dual-threat) often rely on subjective observation. Recent work using rushing contribution percentages and machine learning has shown clear statistical groupings. However, few studies have incorporated CPOE — an advanced Next Gen Stats metric that adjusts completion percentage for depth of target, pressure, and situational difficulty.This study aims to:

1. Apply PCA + clustering with CPOE to define objective archetypes.

2. Compare fantasy production across groups.

3. Provide visual and statistical validation of the three-archetype model (Dual-Threat, Balanced, Pocket).

Methods

Data Sources

2025 regular season statistics for starting and significant backup quarterbacks were compiled, including:

• Passing: Yards per game, Completion %, Touchdowns per game, Interceptions per game

• Rushing: Yards per game

• Advanced: CPOE (Completion % Over Expected)

• Contextual: Age

Feature Engineering

Seven normalized features were used: Pass Yds/G, Comp%, Pass TD/G, INT/G, Rush Yds/G, Age, and CPOE.Analytical Pipeline

• Data standardization (z-score)

• Principal Component Analysis (PCA) for dimensionality reduction

• K-Means clustering (k=3) on standardized features

• Archetype labeling based on cluster centroids (primarily driven by Rush Yds/G and CPOE)

• Fantasy Points Per Game (FPG) calculated using standard PPR/Superflex scoring

Visualization

PCA scatter plots, boxplots of FPG and CPOE by archetype, and summary statistics were generated.