← Back to Learn

MLB Home Run Props: What Actually Predicts Home Runs

We trained an XGBoost model on 136,000 batter-games across three MLB seasons. Here is what actually predicts who will hit a home run tonight — and why barrel rate, weather, and the pitcher matchup matter more than season HR totals.

Published April 2026 · 15 min read

1. The Home Run Betting Market

Home run props are the most exciting bet in baseball. A single HR can flip a game, and the moment the ball clears the fence is immediate, unambiguous, and completely detached from the final score. You do not need your team to win. You do not need a bullpen to hold a lead. You just need a batter to make one perfect swing.

That simplicity has made HR parlays enormously popular — and occasionally enormous. A $30 parlay linking four to six home run props in a single slate recently paid $1.83 million. These payouts are real, and they are not flukes. They happen because HR props compound well: at +300 each, a four-leg parlay pays over 80-to-1, meaning a single correct $30 ticket covers years of losing bets if you are picking the right legs.

But HRs are rare. Even Aaron Judge, one of the most prolific power hitters in modern baseball history, homers in only about 35% of his games at peak production. The league average is approximately 11% — meaning roughly 89 out of every 100 batter-game appearances produce no home run at all. The overwhelming base rate is zero.

This is why naive approaches to HR props fail. You cannot simply list the most powerful hitters in baseball and bet them every night. Picking Aaron Judge every game at +300 is not a winning strategy because he does not homer in 65% of games, and you will lose that +300 bet twice for every time you win it.

The correct framing is not: "Will this batter hit a home run tonight?" It is: "Which batter tonight has the highest probability of hitting a home run — and is that probability being underpriced by the sportsbook?" That is a ranking problem. And ranking problems are exactly what machine learning models are built to solve.

2. Why Most HR Predictions Fail

The failure mode in most public HR prediction approaches is not a lack of effort — it is a structural mismatch between the model type and the problem.

The base rate problem is severe. When 89% of outcomes are zero, any binary classifier trained to predict "will this batter homer?" will learn very quickly that predicting "no" every time achieves 89% accuracy. The model learns the base rate, not the signal. This is why simple binary classifiers trained on batter-game data produce outputs that look accurate in aggregate but are useless for betting: they predict "no HR" for almost every batter, which is often technically correct and always practically worthless.

Season-level statistics compound the problem. Career HR rate, season HR per plate appearance, and lifetime home run totals are all real signals — but they miss the game-to-game variance that determines whether tonight is one of the 11% or one of the 89%. A batter with a .250 season ISO who has gone 0-for-3 in HRs over his last 15 games is in a different situation than the same batter who has homered in three of his last seven. The season-level average tells you nothing about which situation you are in.

What Shifts the Probability Game-to-Game

Several factors shift HR probability significantly on a game-by-game basis, none of which are captured by season-level stats:

Recent formA batter's rolling batting average and contact quality over the last 5-10 games. Hot streaks are real and measurable — see Section 3.
Pitcher matchupHigh fastball percentage pitchers allow more HRs. A high-barrel batter facing a HR-prone pitcher is a multiplicatively different situation.
WeatherWind direction and speed, temperature, and humidity all affect carry distance. Our model assigns 18.2% feature importance to weather variables.
ParkCoors Field at altitude produces a 1.45x HR factor. Oracle Park suppresses HRs to 0.86x. The same batter is a different proposition in each park.

Most public HR prediction sites rank batters by season HR total or average HR rate. These rankings change very slowly — sometimes not at all from week to week. They completely ignore tonight's specific conditions. That structural gap between what they publish and what actually drives HR probability is where a properly designed model finds its edge.

3. The Hot Hitter Effect — Validated by Data

The "hot hand" is a famously contested concept in sports analytics. In basketball free throws, the evidence is mixed. In baseball hitting — specifically home run production — it is real, measurable, and has material betting implications.

We analyzed 129,000 batter-games across three MLB seasons and found a strong, consistent relationship between recent hitting performance and same-night HR probability. The findings are not subtle:

1

2x HR rate when hitting .300+ over last 5 games

Batters hitting .300 or better over their previous five games hit home runs at twice the rate of batters hitting below .150 over the same window. This is not a small sample artifact — it holds consistently across three seasons and 129,000 batter-games. Recent contact quality is a genuine leading indicator of same-night HR probability.

2

46.7% of all HRs come from batters on a hot streak

Nearly half of all home runs in the dataset were hit by batters with a rolling .300+ average over their last five games. By contrast, only 17% came from cold hitters (below .200 rolling average). This concentration is the quantitative basis for the hot hitter ranking — finding the batters most likely to be in the high-probability tier requires identifying who is currently hot, not just who has historically been powerful.

3

Rolling ISO elevated +0.201 in 5 games before a HR

Isolated power — slugging percentage minus batting average — is elevated by an average of +0.201 in the five-game window immediately preceding a home run game. Batters are hitting for more extra-base power in the days before they homer. This is not survivorship bias: it holds even when controlling for long-term ISO baseline and suggests that short-term power form is a real and actionable signal.

4

Lower K rate correlates with higher HR probability in hot windows

Contact rate matters within hot streaks. Batters who are making more contact — lower strikeout rate in the recent window — also homer at a slightly higher rate, suggesting they are genuinely "seeing the ball well" rather than simply benefiting from a random cluster of quality swings. This makes mechanical sense: a batter with better contact is squaring up pitches more consistently, which is both why they are making contact and why they are generating better launch conditions on their hard contact.

4. Statcast Features That Drive HR Probability

Our model's top feature by importance is expected isolated power (xISO) from Baseball Savant, at 16.5% of total feature importance. xISO synthesizes barrel rate, exit velocity, and launch angle into a single expected power output metric that predicts extra-base production better than any of its component inputs in isolation.

Here is the full Statcast profile that drives HR prediction:

Barrel Rate

Barrel rate is the percentage of a batter's batted balls that qualify as "barreled" — hit at the optimal combination of exit velocity (95+ mph) and launch angle (25-35 degrees) that maximizes the probability of extra-base hits. The league average barrel rate is approximately 7%. Elite power hitters sit at 15% or higher.

Barrel rate is the most direct measurable proxy for a batter's ability to generate home run contact. A batter with a 14% barrel rate is not twice as likely to homer as a batter with a 7% barrel rate — the relationship is nonlinear and depends on park and weather — but barrel rate is the single best Statcast input for HR prop assessment.

Exit Velocity

Exit velocity measures how hard the ball comes off the bat, in miles per hour. Batters averaging 95+ mph on contact are in the power tier. The relationship between exit velocity and HR probability is not linear — it jumps sharply above 100 mph, where even off-angle contact can clear the fence at hitter-friendly parks.

Average exit velocity is more stable than barrel rate as a long-term indicator, but recent exit velocity trends are a useful signal for detecting batters whose contact quality is improving or declining within the season.

Launch Angle

The optimal home run launch angle is 25-35 degrees. Fly ball hitters with a pull tendency generate more HR opportunities than ground-ball hitters, even at equivalent exit velocities, because ground balls and line drives below 15 degrees almost never result in home runs regardless of how hard they are hit.

Launch angle is where intentional swing changes show up first in the data. When a batter deliberately elevates his swing path to generate more fly balls, launch angle shifts before barrel rate and HR totals catch up. Identifying launch angle improvement is one way the model detects power breakouts before the market prices them in.

Pull Percentage

Pull percentage — the fraction of batted balls hit to the pull side — interacts directly with park factors in a way that matters for HR props. A pull-heavy right-handed hitter at Yankee Stadium, with its famously short 314-foot right porch, is a categorically different bet than the same batter at Oracle Park, where right-center field extends 399 feet.

Our model treats pull percentage as a modifier on top of barrel rate and exit velocity, because the same quality of contact produces different outcomes depending on the direction of the ball and the dimensions of the park. High pull percentage + short porch = elevated HR probability over and above what the Statcast metrics alone would suggest.

See Tonight's HR Rankings

Our XGBoost model runs every day before first pitch, ranking every batter by HR probability using 48 features. Top 5 batters hit HRs at 2.2x the baseline rate across 136,000 backtested batter-games.

5. The Pitcher Matchup

Not all pitchers allow home runs at equal rates, and the differences between them are large enough to materially shift a batter's HR probability on any given night. A power hitter facing a high-fastball pitcher is in a fundamentally different situation than the same batter facing a sinkerballer working the bottom of the zone.

Fastball Percentage and HR Vulnerability

Fastballs are the easiest pitch to drive for power. They travel on a predictable path, arrive earlier in the zone, and give batters more time to generate bat speed through impact. High fastball percentage pitchers — those who throw four-seamers and sinkers on 60%+ of pitches — allow more home runs per nine innings than pitchers who rely heavily on breaking balls and changeups.

The logic is straightforward: power hitters sit fastball. When a pitcher throws fastballs frequently, the power hitter is getting more looks at his best pitch. Hard-throwing fastball pitchers who challenge power hitters — rather than tunneling off-speed pitches below the zone — are the most exploitable matchups for HR props.

Rolling Pitcher HR Vulnerability

Our model uses barrel rate allowed and HR per nine innings from the opposing pitcher's recent starts — specifically his last seven outings — rather than his full season average. This is an important distinction.

A pitcher who has allowed five home runs across his last three starts is more exploitable today than his season average suggests. He may be tipping pitches, losing command of his secondary offerings, or struggling mechanically in a way that will take two to three weeks to show up in his season-level stats. Rolling pitcher HR vulnerability is a leading indicator that the market often prices too slowly.

Conversely, a pitcher with a strong season HR/9 who has tightened up recently — fewer hard-hit balls allowed, better command of his pitch mix — is being undervalued by bettors anchored to his season line. Rolling windows cut both ways.

The Batter-Pitcher Interaction

The interaction between batter profile and pitcher profile is what the model captures that simple stat lookups miss. A high-barrel batter facing a high-fastball pitcher is multiplicatively more dangerous than either stat in isolation suggests.

This interaction effect — similar to the matchup_k_potential feature in our strikeout model — is one of the highest-importance features in the HR model. It captures the idea that the right batter against the right pitcher on the right night is a qualitatively different bet than the same batter on an average night. Identifying those nights in advance is the model's primary job.

6. Weather: The Hidden Variable

Weather is the most underappreciated input in HR prop betting. Most bettors do not check wind conditions before placing HR bets. Most public models do not incorporate weather at all. Our model assigns weather variables 18.2% of total feature importance — more than any single Statcast metric except xISO. This is the gap.

Wind

Wind blowing out at 10 miles per hour or more is the single largest weather factor in HR probability. At Wrigley Field, when the wind blows out to center or left-center, the over hits on totals at 60%+ and individual HR props see a measurable boost. A 10 mph wind carries a well-struck ball roughly 15-20 additional feet — the difference between a warning track out and a home run at many parks.

Wind direction matters more than wind speed at moderate levels. A 15 mph wind blowing in from center suppresses HRs more than a 20 mph crosswind. Our model incorporates both speed and direction as separate inputs because their effect on ball carry is not equivalent across parks with different orientations.

Temperature

Every 10 degrees Fahrenheit above 75 degrees adds approximately 3 feet of carry to a well-struck ball. An 85-degree afternoon game versus a 55-degree night game is a measurably different environment for HR production — roughly 9 feet of carry difference on every batted ball. That is the margin between a warning track out and a home run multiple times per game.

Temperature effects are most pronounced at extreme ranges. Games above 90 degrees see noticeably elevated HR rates at outdoor parks. Games below 50 degrees — early April night games in northern cities — suppress HR production meaningfully. The model does not apply a simple linear temperature adjustment; it learns the nonlinear relationship from historical data.

Humidity

Humidity has a counterintuitive effect that most bettors get backwards. Humid air is actually less dense than dry air, because water vapor (H2O, molecular weight 18) is lighter than the nitrogen (28) and oxygen (32) that make up the bulk of dry air. When water vapor displaces heavier gas molecules, overall air density drops slightly — and lower air density means less aerodynamic drag on a batted ball, which means more carry.

The effect is small — a few feet of carry difference between very dry and very humid conditions — but it is real and in the opposite direction from what most people assume. Humid summer games at outdoor parks are slightly more HR-friendly than dry games at equivalent temperature and wind conditions.

Domes and Weather Neutralization

Retractable-roof and fully enclosed domes neutralize all weather effects entirely. At parks like Tropicana Field, Rogers Centre, Chase Field (when closed), and Globe Life Field (when closed), HR probability is driven purely by the batter-pitcher matchup and park dimensions. No wind, no temperature carry, no humidity effect. For dome games, the model weights matchup and Statcast features more heavily and drops weather inputs.

Our weather interaction feature — high barrel rate combined with wind blowing out — produces the maximum HR probability boost in the model. When a 15%+ barrel rate batter faces a high-fastball pitcher on a day with 15+ mph outfield wind, the model's output moves sharply upward from the baseline. These are the highest-conviction HR plays the system generates.

7. Park Factors for Home Runs

Not all parks are equal. The difference between the most HR-friendly park in baseball and the most suppressive is not marginal — it represents a 69% swing in HR probability for identical batted balls. Our model uses HR-specific park factors rather than general run-scoring factors, because a park can suppress overall run scoring while still being relatively HR-friendly (short porches with deep gaps) or vice versa.

ParkHR Factor
Coors Field (COL)1.45x
Yankee Stadium (NYY)1.15x
League Average1.00x
Oracle Park (SF)0.86x

Why Coors Field Is Unique

Coors Field sits at 5,280 feet above sea level. At altitude, air density is measurably lower, which reduces aerodynamic drag on every batted ball. A well-struck fly ball at Coors carries 15-25 feet farther than the same ball at sea level. Combined with Coors's expansive outfield dimensions, this produces the highest HR factor in baseball — 1.45x the league average — despite the park's large outfield surface.

Yankee Stadium's Right Field Porch

Yankee Stadium's 314-foot right field line is the shortest in the American League East and among the shortest in baseball. Right-handed pitchers giving up pull-side fly balls to left-handed batters at Yankee Stadium face a structural disadvantage — balls that would be comfortable outs in most parks become home runs. The park factor of 1.15x reflects this, but the effect is even more pronounced for left-handed power hitters with high pull rates.

Oracle Park's Suppression Effect

Oracle Park (San Francisco) suppresses HRs to 0.86x the league average through a combination of cold temperatures, persistent wind blowing in from McCovey Cove off the Bay, and deep power alleys (399 feet to right-center). A batter who averages 35 HRs per season at a neutral park would project to approximately 30 at Oracle and 40 at Coors under identical conditions. For HR prop purposes, always cross-reference the park before placing a bet on a power hitter.

For a deeper look at how park factors affect both HR props and totals betting, see our MLB park factors and totals betting guide.

8. How Our Model Ranks Batters

Our HR ranking model uses XGBoost gradient boosting with 48 input features to produce a HR probability estimate for every batter in tonight's confirmed lineups. The output is a ranked list — not a binary yes/no prediction — because HR prediction is fundamentally a ranking problem. The goal is to identify the top tier each night, not to certify individual outcomes.

The model was walk-forward backtested across 136,000 batter-games over three MLB seasons. Walk-forward means the model was never allowed to train on future data to evaluate past predictions — every projection is generated using only data available at the time of the game, the same way it runs in production.

CohortHR RateLift vs. Baseline
Top 3 ranked batters23.5%2.1x
Top 5 ranked batters23.8%2.2x
Top 10 ranked batters22.5%2.0x
Shuffle test (random)11.5%baseline
All batter-games (baseline)~11%1.0x

What the Shuffle Test Tells Us

The shuffle test is the most important validation step. We took the same 136,000 batter-games, randomly shuffled the batter identities within each game date, and ran the identical ranking process. The result was 11.5% HR rate for the "top 5" — essentially the baseline rate.

The 11.5-percentage-point gap between the shuffled result (11.5%) and the real model output (23.8%) is the cleanest evidence that the model's feature set contains genuine signal. If the model were overfitting to noise, the shuffle test would produce nearly the same result as the real model. The fact that it collapses to near-baseline confirms that the ranking is driven by real features, not spurious correlations.

What the Model Does Not Do

The model does not predict certainty. A batter ranked #1 tonight is not going to homer — he is going to homer in roughly 23-25% of the nights he occupies that position. Three in four of those nights will produce no home run regardless of how perfectly the model has identified the favorable conditions. This is the irreducible variance of a rare event, and it is not a failure mode. The value is cumulative: over hundreds of bets, the difference between 23% and 11% is enormous.

9. Practical: Using HR Rankings for Betting

Understanding what drives HR probability is useful. Translating it into a practical betting process requires one additional step: comparing the model's implied probability against the sportsbook's implied probability and identifying where the gap is large enough to bet.

Building Your Shortlist

The top 3-5 batters in the daily HR rankings are your shortlist. These are the batters for whom the model has identified the highest probability combination of factors tonight. Start here, not with a list of all power hitters in baseball.

Cross-reference with the weather page at /mlb-weather for current wind conditions at tonight's outdoor parks. A top-5 batter playing at a park with wind blowing out at 12+ mph is a higher-conviction bet than the same ranking on a calm night. A top-5 batter playing in a dome needs to be evaluated purely on the batter-pitcher matchup and park dimensions, with no weather boost.

Single HR Props vs. Parlays

Single HR props — player to hit one or more home runs — are typically priced at +250 to +450 odds, implying a probability between 18% and 29%. At +300 (25% implied), you need a true probability above 25% to have a positive expected value at standard juice.

Our top 5 hit at 23.8% in the backtest — close to the +300 implied probability but not definitively above it on its own. This means single HR props are not slam-dunk positive EV bets on every top-5 batter. The edge is more nuanced: the top-5 ranked batters are priced collectively around +300 by books, but individual batters within that group are not uniformly priced. A top-ranked batter facing a high-fastball pitcher with wind blowing out might be a true +24% probability priced at +350 (+22% implied) — that is a +2% edge. A different top-5 batter in a dome on a poor matchup might be a true +19% priced at +275 (+27% implied) — that is negative EV despite the ranking.

The $1.83M Parlay Approach

Parlaying two to three top-ranked batters is high-risk, high-reward. A three-leg parlay of batters at +300 each pays roughly 80-to-1 if all three homer. With a true 24% probability per leg, the true probability of all three homering is approximately 1.4%. At 80-to-1, that is positive expected value — but with a standard deviation so large that you will lose the vast majority of tickets placed.

The $1.83M parlay payouts are real. They happen because the combinatorial odds on multi-leg HR parlays are enormous relative to the true probability when the legs are correctly identified. The risk is that any single missed leg — and remember, even the #1 ranked batter misses 76% of the time — kills the entire ticket. Parlay betting on HR props is only viable with strict unit sizing and an explicit acceptance of frequent total loss.

Where the Edge Actually Lives

The clearest edge in HR props is not in blindly betting every top-5 batter. It is in identifying which top-5 batters are priced at +EV given their specific combination of factors tonight. A batter ranked #2 facing the most HR-vulnerable pitcher in baseball with wind blowing out 15 mph might be priced at +350 by a book still anchored to his season HR rate. The model sees a true probability closer to +28%. That gap — not the ranking itself — is where the edge lives.

For a broader look at how to use player prop models across all MLB categories, see our MLB player props guide.

Check Tonight's Wind Conditions

Weather accounts for 18.2% of our HR model's feature importance. Before placing HR props, check which outdoor parks have favorable wind conditions tonight.

View MLB Weather & Wind

10. Frequently Asked Questions

How accurate are MLB home run predictions?

Our XGBoost model backtested across 136,000 batter-games over three seasons shows that the top 3 ranked batters hit home runs at 23.5% — 2.1x the 11% baseline rate. A shuffle test produced only 11.5% accuracy, confirming the signal is real and not data leakage. No model predicts which batters will definitely hit a home run on a given night. The value is in consistently identifying which batters are most likely relative to the field, then finding spots where that probability is underpriced by the sportsbook.

What is the best stat for predicting home runs?

Expected isolated power (xISO) from Baseball Savant is the single most important feature in our model at 16.5% importance. xISO combines barrel rate, exit velocity, and launch angle into a single expected power output metric. Barrel rate alone — the percentage of batted balls hit at the optimal launch angle and exit velocity combination — is the most useful single Statcast stat for quick assessments. League average barrel rate is around 7%; elite power hitters sit at 15% or higher.

Does weather affect home runs in baseball?

Yes — significantly. Weather accounts for 18.2% of total feature importance in our HR model. Wind blowing out at 10+ mph is the biggest single weather factor; at Wrigley Field, overs hit 60%+ when wind blows out to center. Temperature adds roughly 3 feet of carry per 10 degrees Fahrenheit above 75. Humidity counterintuitively helps slightly, because humid air is less dense than dry air — water vapor is lighter than nitrogen and oxygen. Domed stadiums neutralize all weather effects entirely.

What is barrel rate and why does it matter for HR props?

Barrel rate is the percentage of a batter's batted balls classified as barreled — hit at the optimal combination of exit velocity (95+ mph) and launch angle (25-35 degrees) that maximizes home run probability. The league average is approximately 7%; elite power hitters like Aaron Judge or Shohei Ohtani barrel 15-20% of their batted balls. For HR props, barrel rate is the most direct measurable proxy for a batter's ability to consistently generate home run contact and is a leading indicator that updates faster than HR totals themselves.

How often do the best power hitters actually hit home runs?

The league-average batter hits a home run in approximately 11% of games. Elite power hitters like Aaron Judge homer in roughly 35% of their games at peak production — more than 3x the baseline. Even the best power hitters fail to homer in 65% of their games. HR props are typically priced at +250 to +450 odds. The betting edge is not in identifying who will definitely homer; it is in identifying which batters at +300 are actually closer to a 24% true probability versus the 11% implied by a naive baseline. That gap between true probability and implied probability is where value lives.

See tonight's HR rankings — all 48 features applied

Our XGBoost HR model runs every day before first pitch, ranking every batter in tonight's lineups by home run probability. Top 5 ranked batters hit at 2.2x the baseline rate across 136,000 backtested batter-games. Compare our rankings against your book's lines before placing HR props.

Start your free 5-day trial — predictionengine.app/pricing

5-day free trial — all sports

Try Free