"Random Forest-Driven Feature Analysis and Composite Scoring for PPR Fantasy Football RB Drafting"

Abstract

Fantasy football drafting benefits from machine learning techniques such as Random Forests (RF) that identify non-linear levers of success. We applied RF-inspired feature importance insights, statistical validation (correlations, OLS regression, ANOVA + Tukey HSD), and a custom composite "PPR RB Levers Score" incorporating volume, receiving work, goal-line usage, yards after contact, consistency, team context, efficiency, and an age-decline penalty (26+). The model was backtested on synthetic and leader-based data, achieving strong predictive power (R² ≈ 0.88 for simulated performance). Results highlight receiving volume and goal-line share as dominant in PPR formats, with clear value identification relative to 2026 ADP. The approach provides a reproducible framework for data-driven drafting.

Methods

Data: 2026 FantasyPros RB ADP rankings (top 72) combined with approximated 2025 season stats grounded in real leader patterns (carries, receptions, YPC, etc.). New levers added: goal-line/red-zone share, yards after contact, and fumble rate. 

Feature Engineering & Normalization: Min-max scaling to 0–100 for all variables. Age score implemented as a linear penalty starting at 26.

Model:

• Composite score: Weighted sum (0.24 rec + 0.18 carries + 0.15 consistency + 0.13 goal-line + 0.10 YAC + 0.09 age + 0.07 team share + 0.04 fumble-adjusted).

• Validation: Pearson correlations, multiple OLS regression (predicting simulated next fpts), and ANOVA/Tukey HSD for tier separation.

Results

The correlation matrix confirmed receiving volume as the strongest lever (r ≈ 0.61 with score). Goal-line share and YAC added independent value (r ≈ 0.37–0.38). Age showed the expected negative relationship.OLS regression on simulated performance yielded R² = 0.883, with significant positive coefficients for carries, receptions, consistency, goal-line, and YAC, and a negative coefficient for age.

See Figure 2: Correlation Heatmap (full matrix of levers, score, and simulated performance)

See Figure 3: ADP vs Levers Score scatter (split top/bottom 36 with all labels) – clear separation of value plays above the trend line.

Top scorers were young, high-reception RBs with strong goal-line usage (e.g., Gibbs, Robinson, Jeanty). Older volume backs often fell below the ADP-expected score.

Discussion

RF-derived levers translate effectively into a practical drafting tool. The model’s emphasis on PPR-specific receiving work and scoring opportunity aligns with real-world analytics (e.g., POP model insights).

 Limitations include reliance on approximated stats and synthetic backtesting — future work should integrate live 2025 data and injury risk. 

The framework is fully reproducible and extensible to other positions. Drafters can use the Levers Score to identify market inefficiencies, particularly undervalued young pass-catching backs.

Bottom Line: On the Record for Beta Testing this concept!