Skip to main content
Retrained nightly on 468 stocks

AI Stock Predictions Powered by
10 Machine Learning Models

Every day, 10 independent ML algorithms analyze every S&P 500 stock and vote on direction. When most models agree with high confidence, you get a clear BUY or SELL signal.

Try Free →

10 Algorithms, One Ensemble Signal

Each model brings a different strength — tree-based models catch non-linear patterns, deep learning models detect sequential trends. The ensemble combines them for higher accuracy.

Random Forest
Tree Ensemble
Gradient Boost
Tree Ensemble
Extra Trees
Tree Ensemble
XGBoost
Gradient Boost
LightGBM
Gradient Boost
Neural Net
Deep Learning
LSTM
Recurrent
GRU
Recurrent
TCN
Temporal CNN
TabNet
Attention

101 Features Per Stock, Every Day

Models don't just look at price. They ingest technical indicators, earnings data, fundamentals, sector momentum, sentiment, and options flow.

61
Technical Indicators

RSI, MACD, Bollinger Bands, ADX, ATR, OBV, VWAP, Stochastic, Ichimoku, and 52 more

40
Enrichment Features

Earnings surprise, P/E, P/B, debt ratio, sector momentum, VIX correlation, options put/call ratio, institutional flows

5
Target Variants

Price up, beat SPY, >0.5% gain, >1% gain, positive + beat index — models train on multiple definitions of "success"

4 AI Advisors for a Second Opinion

Beyond the ML ensemble, four large language models independently analyze each stock. When multiple advisors agree, the conviction is higher.

🟣
Claude

Deep fundamental reasoning about market context, risks, and catalysts.

🔵
Gemini

Broad market perspective with real-time data and trend analysis.

🟠
Llama

Fast portfolio reviews and quick second opinions on positions.

🟢
ML Ensemble

10 algorithms vote independently. Majority + high confidence = strong signal.

How We Measure Prediction Accuracy

Directional accuracy: We measure whether the predicted direction (up or down over 5 trading days) matches the actual price movement. A model that says "up" and the stock goes up is a correct prediction.

Time-series validation: Training data is split chronologically — models train on older data and validate on newer data, never the reverse. This prevents look-ahead bias that inflates accuracy in backtests.

Overfit rejection: Any model achieving >99% AUC on training data is automatically rejected. This catches models that memorize rather than generalize.

Nightly retraining: Models retrain every night with the latest market data. Stale models degrade — fresh models adapt to changing market conditions.

Past performance does not guarantee future results. All predictions are probabilistic, not guaranteed.

Start Getting AI Stock Predictions

Free paper trading included. No credit card required.

Get Started Free →