Recently, everyone's been talking about what's happening in Iran, but there's a whale on Polymarket who quietly made over $165,000 (roughly 1.2 million RMB) from the tennis market without making a sound.

This person never attends matches in person or messages coaches. What he relies on is a machine learning model running on his computer.

Here's his wallet if you want to check it out:

Looking at his two recent plays, they're pretty slick:

Paris Masters: Sinner vs. Bonzi
Profit: $25,184.10 (+47.55%)

Paris Masters: Christian vs. Tian
Profit: $24,413.96 (+166.7%)

How does he dominate tennis betting? Basically, four steps.

1. Built a "data holy grail"
He fed the model nearly 100,000 professional matches from 1985 to 2024. Court type, service faults, break points—everything you can think of.

But raw data alone isn't enough. Here's what made him really smart—he calculated:
The win rate differential between two players
The age gap
Their ELO ratings on specific surfaces (clay, grass, etc.)

2. Identified the critical data points
Just like on the Titanic, "first class" and "female" were the key survival indicators. He excavated through massive datasets and isolated two factors that best predict wins:
Overall strength differential between the two players
Their strength differential on today's specific surface

Math proved something simple: if you're facing Nadal on clay, his "clay court ELO halo" is basically unbeatable.

3. Got the models to work in relay
He started with a single decision tree—74% accuracy. Then tried basic ELO rules—72%. Then deployed a "random forest" model (like 94 trees voting together)—accuracy hit 76%, still not enough.

Finally, he deployed the heavy weapon: XGBoost.

This isn't about trees voting together; it's about trees working in relay. Each tree focuses on where the previous tree got it wrong and patches the gaps. Combined with "regularization" techniques to prevent overfitting, accuracy jumped to 85%, even beating more complex neural networks.

4. Validated with real-world testing
He trained the model using data up to 2024, then predicted the 2025 Australian Open right after it finished.

Results:
116 matches, got 99 right (85.3% accuracy)
Before matches even started, the model predicted Sinner would go undefeated to win.

No insider information whatsoever. Just:
One computer
Open-source Python code
XGBoost algorithm
And the guts to place big bets on a market not yet saturated with attention.

I'm planning to follow along myself.

There's a copy trading bot—connect your wallet and it automatically mirrors his positions.

Click here to start syncing his portfolio: