NBA Player Proposition Markets: A Two-Tier Machine Learning System with Five-Month Out-of-Sample Validation

This paper presents a validated two-tier betting system for NBA player proposition markets, developed using XGBoost, Random Forest and GLMNet models trained independently per market. A five-month walk-forward out-of-sample test across 13 markets confirmed both tiers independently profitable every month. Tier 1 (model probability >= 0.80) achieved 16.0% average monthly ROI from a $2,000 starting bankroll, reaching $12,290. Tier 2 value bets (probability 0.60-0.79) achieved 28.2% average monthly ROI, reaching $23,321. Kelly criterion sizing with per-bet and daily caps is detailed. Markets with negative out-of-sample performance were permanently excluded. Three-gate validation criteria ensured zero selection bias and zero data leakage. The system is implemented at betbetter.world.