About — Oddsix

The mission

Why I built this

I built Oddsix as a personal project, driven by a deep passion for the NBA. I've been watching basketball for years and wanted to build something that combined that love for the game with the skills I've been developing in data science and machine learning.

At its core, this started simply — I wanted a place where I could dig deep into player stats, compare players across different eras, and spend my free time doing something I genuinely enjoyed. Over time it grew into something more: a real machine learning pipeline, game outcome predictions, and a tool that explains every recommendation in plain English — so anyone can understand why a pick might have value, not just that someone thinks it does.

This is a showcase of what I've learned and built. It's not perfect, but it's real — real data, real models, real predictions. And it's something I actually use.

— Spencer, creator of Oddsix

The process

How it works

STEP 01

We collect the data

Every NBA game since 2000. Box scores, lineups, injuries, betting lines. Updated hourly during the season.

STEP 02

Our model finds edges

A machine learning model trained on 390+ features — including Elo ratings, Four Factors composites, EWMA trends, and cross-matchup interactions — identifies when sportsbooks misprice a game. 68.2% accuracy on unseen data.

STEP 03

We explain it plainly

Every pick comes with plain-English reasoning. No jargon. No black boxes. You see exactly what the model sees.

What's new

Platform features

Elo Timeline Explorer

Track every team's Elo rating across the full season. Visualize momentum shifts, breakout runs, and collapses on an interactive timeline.

Game Detail + "Why This Pick?"

Click any game card to drill down into a full prediction breakdown. See which features drove the pick, with plain-English explainability.

Player Search + Autocomplete

Find any player instantly with live search and autocomplete. Compare stats, per-game averages, and season trends.

Pinnacle Odds Integration

Live spreads, moneylines, and totals from Pinnacle's sharp market. See implied probabilities alongside model predictions for instant edge detection.

Four Factors Composite

Dean Oliver's Four Factors (eFG%, TOV%, ORB%, FT rate) weighted into a single composite feature. Captures offensive and defensive efficiency in one number.

Interactive Game Cards

Every game card is clickable with drill-down modals showing matchup context, head-to-head records, injury impacts, and model confidence breakdowns.

By the numbers

What's under the hood

68.2%

Model accuracy
game outcome

55.2%

ATS model
accuracy

390+

ML features
engineered

25+

Years of
NBA data

2,140

Tests
passing

3

ML models
in ensemble

See live predictions on the dashboard

Technical details

The models

Game Outcome Model

5-way StackedBlend (GBM 55%/LR 10%/MLP 5%/Margin 30%) — 75 features selected from 390+ (Elo, Four Factors, EWMA, streaks, interactions) — expanding-window cross-validation — 68.2% accuracy, AUC 0.7469

ATS (Against the Spread) Model

Logistic Regression (L1) — Brier-score optimized selection — 54.9% accuracy, AUC 0.5571

Margin Regression Model

Ridge Regression — predicts point margin — CV MAE 10.52, test MAE 10.66

NBAEnsemble

3-model blend with confidence-dependent weights — combines game outcome (0.65), ATS, and margin models — dynamic weighting by prediction confidence

Probability Calibration

Auto-selected (Platt vs Isotonic) — fitted on held-out 2021-22 season — probabilities mapped to [0, 1] true frequencies

Leakage Prevention

shift(1) applied to all rolling features — row N sees only rows 0..N-1 — no same-game information ever used in training

Access

Free forever (for now)

Oddsix is completely free right now. I'm building something I'd want to use myself, and I want to get feedback before charging for anything.

Premium features are coming — but the core data, player comparisons, and model predictions will always have a free tier. Join the waitlist to be first to know when premium features launch.

Go to the dashboard

Legal

Disclaimer: This tool is for informational and entertainment purposes only. It is not financial advice. Past model accuracy does not guarantee future results. Always gamble responsibly and only with money you can afford to lose. Check that sports betting is legal where you are.