• HISTORICAL DATA FROM RETROSHEET (1960-PRESENT)

• DAILY TEAM AND PLAYER PERFORMANCE METRICS

• ADVANCED STATISTICS (wRC, wRAA, BABIP, FIP)

• ROLLING AVERAGES ACROSS MULTIPLE TIME WINDOWS

MONEYLINE MODEL: POWERED BY LIGHTGBM

A GRADIENT BOOSTING FRAMEWORK OPTIMIZED FOR

ACCURACY AND COMPUTATIONAL EFFICIENCY

TOTALS MODEL: POWERED BY NGBOOST

NATURAL GRADIENT BOOSTING TO GENERATE

PROBABILITY DISTRIBUTIONS OF RUNS SCORED

• MARKET ODDS FROM MULTIPLE SOURCES

• TESTED WITH HISTORICAL DATA FROM 2010 ONWARD

• EDGE CALCULATION: MODEL % VS VEGAS IMPLIED %

• THRESHOLD TESTING AGAINST PAST SEASONS

• PERFORMANCE DATA COLLECTED FROM PREVIOUS DAY

• SCHEDULE IS LOADED AND LINEUPS ARE EXTRACTED

• TEAM AND INDIVIDUAL ROLLING AVERAGES CALCULATED

• FRESH ODDS DATA GATHERED

• PREDICTIONS GENERATED AND PUBLISHED AN HOUR OR SO BEFORE FIRST PITCH