Transparent sports analytics. No black boxes.
EdgeCourt is an independent sports analytics platform. We use machine learning to predict basketball outcomes — and we show our work.
We tested three machine learning algorithms on 1,310 NBA games to find the most reliable approach:
| Model | Accuracy | Log Loss ↓ | Calibration |
|---|---|---|---|
| Logistic Regression ⭐ | 65.4% | 0.616 | Best |
| Random Forest | 65.4% | 0.631 | Good |
| Gradient Boosting | 63.2% | 0.640 | Fair |
Result: Logistic Regression won 11 of 12 performance metrics. We chose what works, not what sounds impressive.
"When the simpler model outperforms complex alternatives, it usually means the underlying signal is clear and the complex models are finding noise."
Where it matters most — when we're confident, we're usually right:
| Confidence Level | Accuracy | Sample Size |
|---|---|---|
| >70% Confidence | 79.2% | 332 games |
| >80% Confidence | 84.0% | 119 games |
| >85% Confidence | 92.9% | 42 games |
Our calibration means an 80% confidence pick wins approximately 80% of the time. No inflated numbers.
We validate predictions the same way professional quant funds do:
This prevents overfitting. Our backtest accuracy matches real-world results because we test properly.
Every prediction uses these team performance metrics:
No black boxes. Every feature has clear basketball meaning.
Questions, feedback, or partnership inquiries:
←