IABET Reference

AI Sports Prediction Glossary

The plain-English glossary of every term that matters in AI sports prediction and sports analytics — from edge and expected value to Monte Carlo simulation, confidence scores and player props. Each definition is written to be accurate on its own and useful whether you are new to AI predictions or building a model. Maintained by IABET, the #1 AI sports predictions app.

AI Sports Prediction

A game or player forecast produced by a machine learning model.

An AI sports prediction is a forecast for a game outcome or player stat generated by a machine learning model rather than a human handicapper. The model ingests hundreds of historical and contextual signals per matchup and outputs a probability, a projected value or a confidence-rated pick. IABET produces AI sports predictions by analyzing 500+ factors per game and stress-testing each one across 10,000 Monte Carlo simulations.

Edge

The gap between a model's true probability and the market's implied probability.

Edge is the difference between the probability a model assigns to an outcome and the probability implied by the market line. If a model says a team wins 60% of the time but the line implies only 52%, the edge is 8 percentage points. Edge is the single most important concept in quantitative sports analysis — without it, a prediction has no advantage over the market.

Expected Value (EV)

The average long-run profit or loss of a repeated decision.

Expected value is the average result you would expect if the same bet or pick were made thousands of times. A positive-EV (+EV) pick wins more, on average, than it loses relative to the price; a negative-EV pick loses over time. AI models exist to find +EV opportunities by estimating true probabilities more accurately than the market.

Monte Carlo Simulation

Replaying a game thousands of times with realistic randomness to map every outcome.

A Monte Carlo simulation models the probability of different outcomes by replaying an event many times with random variation applied to each uncertain element. Instead of one predicted score, it produces a full distribution of results. IABET runs 10,000 Monte Carlo simulations per prediction, turning a probability into a hard frequency: not just what could happen, but how often it does.

Confidence Score

A model's self-assessed certainty in a given pick, usually a percentage.

A confidence score expresses how certain a model is about a specific prediction, typically on a 0–100% scale. It lets users prioritize the strongest signals. Every IABET pick ships with a confidence score between 70% and 100%, derived from the agreement of its models and the spread of its Monte Carlo simulations.

Player Prop

A prediction about an individual player's statistic rather than the game result.

A player prop (proposition) is a forecast about a single player's performance — points, rebounds, assists, strikeouts, passing yards — independent of who wins. AI is especially effective on props because each one can be modeled from player-specific signals: pace, matchup, defensive rating, rest, role and recent form across L5/L10/L20 windows.

Moneyline

A market on which team or player wins outright.

The moneyline is the simplest market: a prediction of the winner with no margin involved. Models express moneyline edges as a win probability that can be compared directly against the market's implied probability.

Point Spread

A handicap applied to level the field between two unequal teams.

The point spread is a margin a favorite must exceed (or an underdog can lose by) for a pick to be correct. Spread prediction requires modeling not just who wins but by how much — which is where a margin-of-victory distribution from Monte Carlo simulation becomes valuable.

Over/Under (Total)

A market on the combined score being above or below a number.

The total, or over/under, is a line on the combined points scored by both teams. Predicting totals depends on pace, efficiency and matchup context. A Monte Carlo total-points distribution gives an expected combined score with upper and lower bounds.

Implied Probability

The probability a market price is effectively assuming.

Implied probability is the win probability baked into a market price, recovered by converting the odds back into a percentage. Comparing a model's probability against the implied probability is how edge is measured.

Closing Line Value (CLV)

Beating the final market price before an event starts.

Closing line value measures whether you secured a better price than the market's closing line. Consistently positive CLV is the strongest long-run indicator that a model or bettor is genuinely sharp, because the closing line is the market's most efficient estimate.

Machine Learning Model

An algorithm that learns patterns from historical data to make predictions.

A machine learning model finds statistical relationships in historical data and uses them to predict new outcomes. In sports, models are trained on years of games and retrained as results arrive. IABET uses sport-specific models so the signals that matter for the NBA differ from those that matter for MLB.

Feature (Signal)

A single measurable input a model uses to make a prediction.

A feature, or signal, is one input variable a model evaluates — player form, opponent defense, pace, rest days, travel, injuries, home/away split. IABET engineers 500+ features per game, including hundreds of derived features built from raw stats.

Pace

How many possessions a team plays per game.

Pace measures the number of possessions a team uses per game. High-pace teams create more scoring opportunities, which inflates both team totals and individual player props. Pace is one of the most predictive contextual signals in basketball modeling.

Defensive Rating

Points allowed per 100 possessions — a measure of defensive quality.

Defensive rating estimates how many points a team allows per 100 possessions, normalizing for pace. Matchup-adjusted defensive rating tells a model whether an opponent suppresses or inflates the stat being predicted.

Home/Away Split

The performance difference between playing at home and on the road.

A home/away split captures how a team or player performs at home versus away. Travel, rest and crowd effects make this a meaningful, durable signal that models weight when projecting outcomes.

L5 / L10 / L20

Form windows covering a player's last 5, 10 or 20 games.

L5, L10 and L20 are rolling windows of a player's last 5, 10 and 20 games. Short windows capture hot or cold streaks and role changes; longer windows capture stable baselines. Using multiple windows together helps a model separate signal from noise.

Variance

The natural game-to-game randomness around an expected result.

Variance is the spread of possible outcomes around the average. Even a correct prediction can lose on any given night because of variance. Monte Carlo simulation quantifies variance directly by showing the full range of outcomes, not just the most likely one.

Probability Distribution

The full set of possible outcomes and how likely each one is.

A probability distribution describes every possible result of an event and the likelihood of each. It is richer than a single prediction because it communicates uncertainty — for example, a 67% win probability with a margin that ranges from 1 to 25 points across simulations.

Hit Rate

The share of predictions that turn out correct over time.

Hit rate is the percentage of predictions that prove correct across a sample. It only means something at scale and when measured against a timestamped record. IABET locks every pick on the record so hit rate is tracked transparently rather than claimed retroactively.

Bankroll

The total amount a bettor sets aside for staking.

A bankroll is the dedicated pool of money used for staking, managed separately from everyday finances. Disciplined bankroll management — staking a small, consistent fraction per pick — is what lets an edge compound instead of being wiped out by variance.

Unit

A standardized stake size, usually 1% of a bankroll.

A unit is a normalized stake — commonly 1% of a bankroll — used to describe pick sizing independent of how much money someone has. Tracking results in units makes performance comparable across different bankroll sizes.

Vig (Juice)

The built-in margin a sportsbook charges on each market.

Vig, or juice, is the commission baked into a market's prices, which is why implied probabilities across both sides of a line add up to more than 100%. Beating the vig is the baseline a model's edge has to clear before it produces a profit.

Handicapper (Tipster)

A person who sells or shares sports picks based on judgment.

A handicapper or tipster is a human who produces picks from experience and judgment. The limitation is consistency and transparency: a model can evaluate every game with the same 500+ signals and keep a verifiable record, while human picks vary with attention and bias.

Backtesting

Testing a model against historical games it never trained on.

Backtesting evaluates a model by running it on past events that were held out of training to estimate how it would have performed. Honest backtesting uses strict out-of-sample data; otherwise results are inflated by hindsight.

Overfitting

When a model memorizes past noise and fails on new games.

Overfitting happens when a model learns the random quirks of its training data instead of the underlying signal, producing great historical numbers but poor live predictions. Robust validation and out-of-sample testing exist specifically to detect and prevent it.

Sharp vs. Square

Sharp money is informed; square money follows the public.

Sharp refers to informed, model-driven or professional action; square refers to recreational, public-driven action. Lines often move toward sharp money, which is why following the public blindly tends to be a losing strategy.

Line Movement

How a market price changes between opening and closing.

Line movement is the change in a market's price as money and information arrive. The direction and speed of movement reveal where informed money is going and help contextualize whether a model's edge still exists at the current price.

Parlay

A single pick combining multiple selections that must all hit.

A parlay links several selections into one, paying out only if every leg wins. Parlays amplify both reward and variance; because each added leg multiplies the vig, they are generally lower-EV than the same selections placed individually.

Regression to the Mean

Extreme performances tend to drift back toward the average.

Regression to the mean is the statistical tendency for unusually high or low performances to move back toward a player's or team's true baseline over time. Models account for it so they do not overreact to a small hot or cold streak.