V5 Machine Learning Model

XGBoost gradient boosting with comprehensive feature engineering

~240 FeaturesRolling StatsSituational ContextPhysical MatchupsPlay Style Data

Model Training In Progress

V5 models are currently being trained. Showing placeholder data below.

Model Overview Feature Importance Situational Analysis Prop Performance

Training Samples

~75K

3 seasons of data

Total Features

239

after encoding

Avg Val MAE

4.8

points prediction

vs Baseline

+14%

improvement

How V5 Works

Feature Engineering

Rolling Stats (L5, L10, L15, Season) - Recent form and consistency
Situational Context - Opponent defense, pace, rest days
Matchup Features - Historical vs team, physical differentials
Advanced Stats - Synergy play types, hustle, clutch performance

Model Architecture

Algorithm - XGBoost Gradient Boosting
Separate Models - One per stat type (pts, reb, ast, etc.)
Time-Based Split - Train on historical, validate on recent
Early Stopping - Prevents overfitting

Performance by Stat Type

Stat	Val MAE	Val RMSE	Val R²	Baseline MAE	Improvement	Status
pts	4.80	6.40	0.650	5.60	+14.3%	Strong
reb	2.10	2.80	0.580	2.40	+12.5%	Good
ast	1.80	2.50	0.620	2.10	+14.3%	Strong
stl	0.70	0.90	0.350	0.80	+12.5%	Weak
blk	0.60	0.80	0.400	0.70	+14.3%	Good
tov	1.00	1.40	0.450	1.20	+16.7%	Good
fg3m	0.90	1.30	0.500	1.10	+18.2%	Good

MAE = Mean Absolute Error (avg points off). RMSE = Root Mean Square Error (penalizes large errors). R² = Coefficient of determination (1.0 = perfect).

V5 vs V3 vs V4

Head-to-head comparison once V5 predictions are generated

Baseline Model

Rolling averages + simple factors

Bayesian Model

Probability distributions + regime detection

ML Model

XGBoost + comprehensive features