Education

Learn Crypto Signals

How AI prediction models work, why accuracy ranking matters, and everything we've learned building a platform that holds its own models accountable.

📡

What Are AI Crypto Signals?

BUY, HOLD, SELL — what they mean, how they're generated, and what confidence scores tell you.

🤖

How Our 20 AI Models Work

From XGBoost to Neural Nets — the model zoo, what each type does, and why diversity matters.

🏆

The Ranking Engine

How we score accuracy, weight models, and auto-select the best predictor for each asset.

🔍

Source & Influencer Scoring

How we track 50+ analysts and rank them by actual prediction accuracy — not followers.

📊

How to Read a Signal Card

Confidence bars, model attribution, source consensus — what every element means.

⚠️

Common Mistakes to Avoid

Lessons we learned (the hard way) about data integrity, signal noise, and over-fitting.

Chapter 1

What Are AI Crypto Signals?

An AI crypto signal is a BUY, HOLD, or SELL recommendation generated by a machine learning model. The model analyses price data, volume, technical indicators, and market conditions to predict the probable short-term direction of a cryptocurrency.

Every signal on ArtinFox comes with a confidence score (0–100%). This represents how strongly the ensemble of models agrees on the direction. A 75% BUY means the weighted majority of top-performing models predict the price will go up, with 75% agreement.

Key insight: A signal is only as good as the model that generated it. This is why we don't just generate signals — we continuously rank the models that produce them. A BUY signal from a model with 65% historical accuracy is very different from one with 45% accuracy.

What signals are NOT

Chapter 2

How Our 20 AI Models Work

ArtinFox doesn't rely on a single prediction model. We run a diverse zoo of 20 models, each built on a different analytical approach. This diversity is intentional — different market conditions favour different strategies.

Model Categories

Technical Analysis

  • Technical Indicator — RSI, MACD, Bollinger Band signals
  • Momentum — Rate-of-change and trend strength
  • Mean Reversion — Overbought/oversold detection
  • Volatility Breakout — Bollinger squeeze + expansion
  • Volume Profile — Volume-weighted price analysis
  • EMA Divergence — Exponential moving average crossovers
  • Ichimoku — Cloud-based trend system
  • Parabolic SAR — Trend reversal detection
  • Supertrend — ATR-based directional indicator
  • VWAP Premium — Volume-weighted average price deviations

Machine Learning

  • XGBoost Features — Gradient boosted decision trees
  • Random Forest — Ensemble of decision trees
  • Logistic Regression — Linear probability model
  • Gradient Boosting — Sequential error correction
  • SVM Classifier — Support vector machine boundaries
  • Neural Net — Deep learning pattern recognition
  • Williams %R — Momentum oscillator model
  • Elder Ray — Bull/bear power indicator
  • TRIX Signal — Triple-smoothed momentum
  • Price Action — Candlestick pattern recognition
Why 20 models? Crypto doesn't have a single "best" strategy. Bitcoin in a bull run rewards momentum models. Ethereum during consolidation favours mean reversion. Meme coins are driven by sentiment. By running all 20 simultaneously and ranking them per asset, the platform adapts to whatever the market is doing — without requiring manual intervention.
Chapter 3

The Ranking Engine

Every prediction is logged at the exact moment it's made, together with the closing price at the time. 7 days later, the system checks: did the price move in the predicted direction? This binary outcome (correct / wrong) feeds into the model's accuracy score.

How Accuracy Is Calculated

  1. Daily predictions — Every model generates a BUY/HOLD/SELL signal for every tracked asset, every day.
  2. 7-day resolution — After 7 days, the prediction is resolved against the actual price movement.
  3. Rolling windows — Accuracy is measured over 30-day, 90-day, and 1-year rolling windows.
  4. Per-asset scoring — A model might be excellent for BTC but poor for DOGE. We track this separately.
  5. Weight assignment — Higher-accuracy models receive more influence in the ensemble signal.

The Ensemble Signal

The final BUY/HOLD/SELL you see is not a single model's opinion. It's a weighted vote across all 20 models, where the weight of each model is proportional to its recent accuracy for that specific asset. Models that have been consistently right get more say. Models that have been wrong get dampened automatically.

Lesson learned: Early on, we weighted all models equally. The result was noisy — a bad model could cancel out a good one. Moving to accuracy-weighted voting immediately improved ensemble performance. The best model's signal matters more, and that's reflected in the confidence score.

Market Regime Awareness

The ranker also considers the current market regime. In a strong bear market (BTC trending below its 50-day SMA with negative 7-day and 30-day returns), the system applies a contrarian adjustment — BUY signals need higher conviction to pass through, and SELL signals are weighted more heavily. This prevents the platform from being blindly bullish during downtrends.

Chapter 4

Source & Influencer Scoring

AI models aren't the only opinion that matters. Twitter analysts, YouTubers, research firms, and newsletters all make crypto predictions daily. But how often are they right?

ArtinFox tracks over 50 external sources — from major crypto Twitter accounts to research firms and newsletters. Every prediction they make is scraped, categorised (BUY/SELL), and scored against actual price outcomes.

Position-Based Scoring

Rather than scoring individual tweets (which would be noisy), we use a position-based system:

Conviction Score

We also compute a conviction score for each source — how often do they flip their position within 24–48 hours? Sources that flip rapidly (calling BUY then SELL the next day) get penalised. Consistent, well-reasoned positions score higher. This helps separate genuine analysts from noise machines.

What we discovered: Follower count has almost zero correlation with prediction accuracy. Some accounts with 500k+ followers are below 50% accuracy. Some smaller, focused analysts consistently beat the crowd. The leaderboard reveals this.
Chapter 5

How to Read a Signal Card

Every signal card on ArtinFox packs a lot of information. Here's what each element means:

BUY
Signal Badge
The ensemble recommendation. Based on weighted voting across all 20 models for this specific asset.
72%
Confidence Score
How strongly the models agree. 50% = split vote. 80%+ = high conviction. Below 55% may produce a HOLD instead.
Top 3
Models
Model Attribution
The 3 highest-weighted models driving this call. Lets you see if it's a technical signal, ML signal, or both.
Sources
5/7 BUY
Source Consensus
How many external sources (influencers, analysts) agree with the AI signal. Higher agreement = additional confirmation.
Chapter 6

Common Mistakes to Avoid

We've built and maintained this platform through multiple market cycles. Here are the hardest lessons we've learned — and the mistakes you should avoid when using AI signals.

❌ Treating signals as certainty

Even our best model peaks around 65–70% accuracy over 90 days. That means 30–35% of the time, it's wrong. Never bet your portfolio on a single signal. Use signals as one input among many — alongside your own analysis, risk tolerance, and position sizing.

❌ Ignoring model accuracy windows

A model might show 72% accuracy on a 30-day window but only 55% over 90 days. The shorter window can be misleading — it might reflect a lucky streak during one market condition. Always check the longer windows too. Consistency matters more than peak performance.

❌ Following the hype, not the data

When a popular influencer says "BUY NOW", check their accuracy score on our source leaderboard first. We've seen accounts with millions of followers maintain below-50% accuracy. Popularity ≠ reliability. The data tells you who's actually been right.

❌ Not accounting for market regime

A BUY signal during a strong downtrend carries more risk than the same signal during an uptrend. Our Fear & Greed Index (visible on every page) and the market regime detection help, but always consider the broader macro environment before acting.

✅ What to do instead

Use the model leaderboard to understand which models are performing well right now. Check the source leaderboard to see if influencer sentiment aligns with AI predictions. Review the track record for overall platform performance. Set custom alert thresholds so you're only notified for high-confidence signals. And always — do your own research.

Reference

Glossary

Ensemble Signal
The final BUY/HOLD/SELL recommendation, computed from a weighted vote of all 20 AI models.
Confidence Score
0–100% measure of how strongly the models agree. Higher = more models voting the same way with higher weights.
Directional Accuracy
Did the model correctly predict up or down? BUY + price went up = correct. SELL + price went down = correct.
Resolution Window
The time after a prediction is made before checking the outcome. ArtinFox uses a 7-day horizon.
Market Regime
Whether the market is in a bull, bear, or sideways phase. Determined by BTC's position relative to its 50-day SMA.
Conviction Score
For external sources: how consistently they hold their position vs. rapidly flipping. Higher = more stable thesis.
Fear & Greed Index
0–100 score computed from signal distribution and confidence levels. Extreme fear often precedes bounces; extreme greed often precedes corrections.
Whale Transaction
A large crypto transfer (>$500k USD) between wallets. Exchange inflows may indicate selling pressure; outflows may suggest accumulation.
Ready to explore?

See ranked AI signals in action.

Start with 5 free assets. Upgrade when the data makes you a smarter trader.

Signals are for educational purposes only and do not constitute financial advice. Past accuracy does not guarantee future performance. All investing involves risk.