How Our Prediction Model Works: 16 Weighted Features Explained
By Verdexed Analytics
Every prediction on Verdexed starts with the same core engine: a weighted feature model that evaluates 16+ factors for each game. Unlike talking-head predictions based on gut feelings, our system is entirely quantitative.
The Core Algorithm
For each matchup, we compute a set of features — each scored on a 0-1 scale where values above 0.5 favor the home team. These features are then combined using calibrated weights that reflect each factor's predictive importance.
Sport-Specific Features
**MLB** emphasizes pitcher matchups (12% weight), recent form (12%), and Statcast run expectancy via xwOBA (10%). We also model umpire strike zone tendencies, bullpen strength, and travel fatigue using haversine distance calculations.
**NBA** weights star player impact (10%), defensive rating (10%), and net rating (9%) heavily. Back-to-back detection is critical — teams on zero rest days see significant performance drops.
**NFL** puts the most weight on QB matchup quality (14%), defense strength (10%), and yards differential (8%). We also model weather impact for outdoor stadiums and cold-weather home advantage in late-season games.
**NHL** leverages MoneyPuck expected goals data (xG dominance at 10%), goalie matchup intelligence (12%), and special teams strength. Goalie selection is the single most impactful variable in hockey prediction.
Monte Carlo Simulations
Beyond individual games, we run 10,000 Monte Carlo simulations of the remaining season to project final standings, playoff odds, and championship probabilities. Each simulation uses our game-level win probabilities to randomly resolve every remaining contest.
What Makes Us Different
We don't sell hype. Every prediction comes with a full feature breakdown so you can see exactly what's driving the number. If our model says a team has a 62% chance of winning, you can drill into the 16 factors and see why.