Frequently Asked Questions

TrackBans.com is a machine learning-based system trained on over 200,000 Steam accounts (both VAC banned and clean) to detect behavioral patterns commonly associated with cheaters in Counter-Strike 2. It does not rely on gameplay videos but rather on 180+ public stats, including weapon performance, match history, friends' VAC status, rare skin ownership, and more.

Not in the form of demos or replays. However, it analyzes gameplay data derived from public stats like:

• K/D ratio
• Headshot accuracy
• Weapon-specific performance (AWP, AK, Deagle, etc.)
• Damage, MVPs, round stats
• Match performance metrics
• and more...

In other words: yes, it analyzes gameplay — just not visually.

Cheaters tend to leave behind statistical fingerprints. This AI was trained to learn those patterns — combining dozens of variables (e.g., weapon precision + VAC-banned friends + Steam behavior + rare items) to flag users who resemble known cheaters, even if they're not VAC banned yet.

The percentage represents how closely your profile's patterns match those of accounts known to be cheating.

• Below 50% → Your profile is statistically closer to clean accounts.
• Above 50% → You share behavioral patterns with banned accounts.
• 50% exactly → Neutral. Not enough signal either way.

It's not a ban. It's a red flag.

Yes. It's a beta tool, and like any prediction system, it may produce false positives or false negatives. But with:

• 88.75% recall (detecting actual cheaters)
• 83.97% precision
• ROC-AUC: 0.936

…it's a very solid detection system.

Because the AI doesn't care about what you think you are. It detects statistical similarity. The model has been trained on over 200,000 real Steam accounts. If you're flagged, it means your profile shares traits that have a high correlation with users that got caught cheating.

Some examples:

• You use a small set of weapons with unusually high efficiency
• You have an abnormally high headshot rate
• You play irregularly but with very high performance peaks
• You have multiple friends with VAC bans
• Your inventory or hours played match behavioral profiles of previously banned accounts

These don't make you a cheater, but they raise the risk score. It's pattern recognition at scale.

Not at all. However, some skin patterns are commonly found on cheater accounts, especially bought/stolen accounts. So skins are just one of 180 features — not a deciding factor on their own.

Because that requires access to game files or third-party platforms. TrackBans.com is designed to be:

• Fast
• Public
• Non-invasive
• API-based

And it works in real time, with no downloads or access required.

Yes. The model was trained using Stratified K-Fold Cross Validation (not reusing training data), with real-time testing on unseen samples. It uses:

• RandomForestClassifier with optimized parameters
• StandardScaler to normalize inputs
• Dropout-like strategies (early stopping) to prevent overfitting
• Data from over 200,000 accounts with 180+ features

If that doesn't sound like "real AI" — nothing will.

Because maintaining the servers, protection against DDoS, hosting, storage, and model inference costs real money. TrackBans doesn't sell your data or spam you. Ads help fund the project without charging players.

No — TrackBans.com is an independent project. But its accuracy and predictive power make it a powerful companion to any anti-cheat system. It's a tool to raise awareness and detect risk early.

Soon, a feedback form will be available. For now, remember: being flagged doesn't mean you'll get banned. It's a warning signal, not a punishment.

No. The entire source code is proprietary, but the methodology follows standard and accepted machine learning practices. The confusion matrix proves it isn't guessing.

Because 50% is the model's decision threshold — the statistical "edge" between accounts flagged as clean or suspicious. If you land at exactly 50%, it means your profile has a neutral risk level. This does not mean the system thinks you're cheating — it simply means you don't lean strongly toward either category.

The AI works with real-time Steam API data. If Steam servers are rate-limited, slow to respond, or if some fields time out (inventory, stats, etc.), it may return empty or "hidden" flags — even if your profile is public. In most cases, simply retrying later solves this.

That's partly true — and also true for VAC and Faceit. The AI is trained on real ban data (VAC), so its strongest patterns come from players who were eventually caught. Soft cheaters who never get banned are inherently harder to detect, but the model still picks up suspicious traits and behavior anomalies that often precede bans.

That's how statistics work. 82% accuracy means out of thousands of cases, the AI is correct most times. But it's not perfect and you might be in the 18% of errors. Also, a 50% score is the decision boundary. With 83.97% precision, when the AI flags someone as suspicious (>50%), it's correct about 4 out of 5 times, but 1 in 5 could be wrong.

Absolutely not. The model uses a proper train/validation/test split, which means:

• The data is split into train (60%), validation (20%), and test (20%) sets.
• The model trains on the training set and is evaluated on data it has never seen.
• The final metrics come from the test set — completely unseen data.
• The accuracy numbers (82%+) are from data the model didn't train on.

That's standard practice in real machine learning.

Because cheating isn't binary — it's probabilistic. Some legit players behave like cheaters, and some cheaters mimic legit users. The percentage lets you interpret the result your own way:

• 10%? Chill.
• 60%? Be cautious.
• 85%+? Strong pattern match with known cheaters.

It's up to you to interpret the risk — the system is a detector, not a ban hammer.

TrackBans is a free project, maintained by a single developer. The ads are there to cover server costs, DDoS protection, and infrastructure. There's no paywall, no subscriptions, no data tracking. Just a powerful tool, available to everyone.

No. TrackBans is not a banning system. It's a risk detection tool for personal use, moderation insight, or curiosity. Don't go reporting everyone with 55% and calling them cheaters — that's not the point.