Our Prediction Methodology
Transparency in how we generate data-driven basketball predictions
CourtFrame uses advanced machine learning and statistical analysis to generate basketball game predictions. Our approach combines traditional sports analytics with modern AI techniques to provide data-driven insights.
Important: Our predictions are for informational and entertainment purposes only. Sports outcomes are inherently unpredictable, and past performance does not guarantee future results.
Data Sources
Our predictions are built on comprehensive data from verified sources:
- •Official Game Statistics — Box scores, play-by-play data, and official league statistics from API-Sports and other verified providers.
- •Team Performance Metrics — Win/loss records, home/away splits, recent form, and conference standings.
- •Historical Data — Head-to-head records, seasonal trends, and historical matchup analysis.
- •Advanced Metrics — Offensive/defensive ratings, pace, efficiency metrics, and other advanced statistics.
How Our AI Works
Data Collection & Processing
We continuously collect and process game data, updating our models with the latest statistics. This includes recent game results, injury reports, and team news that may affect performance.
Feature Engineering
Raw data is transformed into meaningful features: rolling averages, momentum indicators, rest days, travel distance, and historical matchup patterns.
Model Prediction
Our ensemble model analyzes all features to generate a prediction. The model outputs both a winner prediction and a confidence score indicating the strength of the prediction.
Human Review
While our system is automated, our team of analysts reviews predictions for major games and can adjust for factors the model might miss, such as breaking news or unusual circumstances.
Understanding Confidence Scores
Each prediction includes a confidence score (0-100%) indicating how strongly our model supports the prediction:
High Confidence
Strong statistical signals supporting the prediction
Moderate Confidence
Mixed signals; closer matchup expected
Low Confidence
Toss-up game; insufficient data for strong prediction
Accuracy Tracking
We believe in transparency. After each game, we track whether our prediction was correct and report our overall accuracy. This helps you understand our track record and make informed decisions.
Historical accuracy statistics are displayed on each prediction and in our monthly reports. We do not hide or delete incorrect predictions.
Limitations
We believe in being upfront about what our predictions can and cannot do:
- •Sports are inherently unpredictable — upsets happen regularly
- •Our model cannot account for last-minute injuries or lineup changes
- •Motivation, team chemistry, and "intangibles" are difficult to quantify
- •Past performance does not guarantee future results
How to Use Our Predictions
Our predictions are designed to enhance your enjoyment of basketball and provide analytical insights:
- • Use them to discover interesting matchups
- • Compare our analysis with your own insights
- • Track accuracy over time to understand model performance
- • Remember that sports outcomes are inherently unpredictable