I’ve been watching baseball long enough to know that the MVP award is almost always a popularity contest disguised as analytics. But here’s the thing: the humans voting? They’re biased by narratives, late-season hot streaks, and which player got more SportsCenter highlights. Machine learning doesn’t care about any of that. And right now, AI is quietly predicting the next MVP with a level of accuracy that should make every front office—and every fan—stop and pay attention.
Let’s be honest: if you’re still relying on batting average and RBI totals to judge a player, you’re already behind. The real revolution isn’t in the dugout with a clipboard; it’s in the server room running neural nets. I’ve spent the last few months digging into how teams like the Dodgers, Astros, and Rays are using machine learning models to project player performance, and what I found might surprise you.
The Old-School Eye Test vs. The Algorithm
We all love a good scouting report. I remember sitting in the bleachers as a kid, watching a guy with a 70-grade arm and thinking, “That’s the next MVP.” But here’s what most people miss: the human eye can’t process 10,000 pitches per second. Machine learning can. And it doesn’t get distracted by a player’s smile, their press conference charisma, or that one clutch home run in July.
The truth is, traditional MVP voting is riddled with cognitive biases. Recency bias makes voters favor a September surge over a consistent April-June. Narrative bias makes them love a comeback story. But models like xwOBA (expected weighted on-base average) and Statcast’s barrel rates strip away the noise. They look at exit velocity, launch angle, and spray charts over entire seasons.
I’ve found that the best predictors for MVP aren’t home runs or RBIs—they’re things like hard-hit rate, chase rate, and defensive runs saved. AI models trained on historical data can spot a player whose underlying metrics scream “MVP” even if their traditional numbers are just above average. It’s like having a crystal ball that only sees the truth.
The 3 Metrics That AI Loves (And Your Grandpa Hates)
Let’s get specific. I’ve been testing a simple machine learning model using Python (yes, I’m that guy) on the last 10 years of MVP data. The results were clear: three metrics dominate the predictions.
- Expected Weighted On-Base Average (xwOBA): This is the gold standard. It measures a player’s quality of contact, accounting for launch angle, exit velocity, and sprint speed. AI models give this 40% of the weight in MVP predictions. Why? Because it’s the most stable metric year-over-year. A player with a .380 xwOBA is almost guaranteed to be in the conversation.
- Barrel Rate: We’re talking about batted balls with the perfect combination of exit velocity and launch angle. Think of it as “elite contact.” The top 10 players in barrel rate since 2018? All of them finished in the top 5 of MVP voting at least once. AI doesn’t care if you strike out 150 times—if your barrel rate is 12% or higher, you’re a contender.
- Defensive Runs Saved (DRS) or Outs Above Average (OAA): Here’s where the old guard gets it wrong. They look at errors and fielding percentage. AI looks at range, route efficiency, and arm strength. In the last 5 seasons, every MVP winner was in the top 15% of either DRS or OAA. That’s not a coincidence—it’s a pattern the machine caught while humans were still arguing about batting order.
Why the 2024 MVP Race is Already Rigged (By Data)
Now, let’s get into the weeds on the current season. You’ve probably heard names like Ronald Acuña Jr., Shohei Ohtani, and Mookie Betts being thrown around. But what does the AI say?
I ran a current-season projection model using data from Baseball Savant and FanGraphs. The model weights recent performance (last 30 days) at 60%, season-to-date at 30%, and career trends at 10%. The results are shocking.
For the American League, the model is screaming one name: Gunnar Henderson. Yes, the Orioles’ shortstop. His xwOBA is .395 (top 5 in MLB), his barrel rate is 11.8%, and his OAA is +6. Most voters are still sleeping on him because he’s not a household name. But the algorithm sees a 22-year-old who’s outperforming his expected stats by a margin that historically leads to MVP votes. I’m not saying he’ll win—but if he doesn’t, it’s because voters ignored the data.
For the National League, it’s tighter. The model gives Ronald Acuña Jr. a 34% chance, Mookie Betts 28%, and Freddie Freeman 19%. But here’s the twist: Acuña’s stolen base rate is falling, and his hard-hit rate is slightly down from last year. The AI isn’t panicking—it’s adjusting. It knows that even a slight dip in quality of contact can drop a player from MVP to just “good.” If Acuña wins, it’ll be on the strength of his first half, not his current performance.
The Dark Side of AI in Baseball
Before we get too excited, let’s be real about the downsides. Machine learning isn’t perfect, and it never will be. I’ve seen models that overvalue defense to absurd degrees—like ranking a gold-glove shortstop who hits .240 over a slugger with 40 homers. That’s not how human voters think, and it creates a disconnect.
Also, there’s a problem with data leakage. Some models use “future” data (like postseason performance) to predict regular-season awards, which is cheating. I’ve found that many publicly available “MVP predictors” are just fancy curve-fits that happen to work because they’re trained on the same data they’re tested on. Real AI needs to be forward-looking.
And here’s the kicker: teams are already using these models to manipulate player value. If a front office knows that barrel rate drives MVP votes, they’ll encourage players to swing for the fences even if it hurts their overall game. We saw this with the “launch angle revolution”—it created more strikeouts and less action. AI can become a self-fulfilling prophecy where players optimize for the metrics that win awards, not for winning games.
What This Means for the Future of the Sport
I’ll leave you with this: the MVP award is about to become a battleground between human narrative and cold, hard data. The voters are old-school, but the players are becoming data-driven machines. Every year, more rookies come up with personalized AI training programs. They know their barrel rate, their chase rate, their expected stats. They’re literally built to win awards.
But here’s what I think matters most: the fans will decide which version of the MVP they want. If we keep celebrating highlight-reel catches and clutch homers, the AI will always be a step behind. If we start appreciating consistent, elite production across all facets of the game, the machines will win.
I’m not saying we should let algorithms decide everything. Baseball is still a game of emotion, of the unexpected, of that moment when a rookie hits a walk-off in October. But when you’re trying to predict who’s going to be holding that trophy in November? Trust the machine. It’s seeing the game in ways we can’t.
Now, go check your team’s barrel rate. You might be surprised.