DocumentationML Predictions
Machine Learning

ML Predictions

Deep dive into our machine learning prediction system and understand how we generate betting insights.

NBA
60.2%
ROI: +9.4%
NFL
57.8%
ROI: +6.2%
MLB
56.1%
ROI: +4.8%
NHL
55.4%
ROI: +3.9%
Prediction Methodology

Machine Learning Models

Our predictions are powered by ensemble machine learning models trained on millions of historical games. We combine gradient boosting, neural networks, and statistical models for robust predictions.

Feature Engineering

Each prediction considers 50+ features including team performance metrics, player statistics, home/away splits, rest days, head-to-head history, and real-time injury reports.

Real-Time Updates

Predictions are recalculated continuously as new data becomes available. Lineup changes, injury reports, and odds movements all trigger model updates.

Edge ML Processing

For instant predictions, we use ONNX Runtime in your browser. This enables sub-second predictions without network latency, perfect for live betting scenarios.

Model Ensemble

We combine multiple model outputs using weighted averaging, where weights are determined by recent performance on similar game types.

Confidence Scores

What is Confidence?

The confidence score (0-100) indicates how certain our model is about a prediction. It's not the win probability, but rather how reliable the prediction is likely to be.

High Confidence (80-100)

Predictions with high confidence have clear patterns in the data and consistent model agreement. These are typically your best betting opportunities.

Medium Confidence (50-79)

Medium confidence predictions show some uncertainty. These can still be valuable but consider smaller bet sizes or combining with other analysis.

Low Confidence (Below 50)

Low confidence indicates conflicting signals or limited data. These predictions should be approached with caution or used for research only.

Confidence vs Probability

A team can have 65% win probability with 90% confidence (we're sure they're favored) or 65% probability with 40% confidence (unclear matchup).

Historical Accuracy

Overall Performance

Our models achieve 58-62% accuracy on moneyline predictions across all sports. While this may seem modest, consistent 58%+ accuracy is profitable long-term.

Sport-Specific Accuracy

Accuracy varies by sport: NBA predictions average 60%, NFL 57%, MLB 56%, and NHL 55%. Basketball's higher sample size and scoring enables better predictions.

ROI Tracking

More important than accuracy is ROI. Our high-confidence predictions (80+) have historically returned 8-12% ROI, outperforming the typical -10% from random betting.

Calibration

Our probabilities are well-calibrated: when we predict 70% win probability, the team wins approximately 70% of the time over large samples.

Transparency

We publish monthly performance reports showing prediction accuracy, ROI by confidence level, and comparison to closing lines. Past performance is available in your dashboard.