Machine learning.
Trained on PSX history.
Four ML models run in parallel on every stock — Linear Regression, Logistic Regression, KNN, and a Neural Network. Their consensus produces a high-conviction directional prediction no single model could match.
Four models. One consensus.
Each model approaches the prediction from a different mathematical angle.
Linear Regression
Supervised Learning
Fits a linear relationship between 20+ technical features and future price movement. Predicts the next-day price target and 5-day projected level.
Logistic Regression
Binary Classification
Classifies each stock as UP or DOWN for the next trading day. Outputs a probability score (0–1) indicating confidence in the directional call.
K-Nearest Neighbours
Instance-Based Learning
Finds the K most similar historical periods in the stock's data and uses their outcomes to predict what will happen next. Excels at recognising repeated patterns.
Neural Network
3-Layer Backpropagation
A 3-layer artificial neural network trained on historical PSX data to capture non-linear relationships between technical features and future price behaviour.
How the ML engine is built
Rigorous feature engineering and consensus scoring — not off-the-shelf signals.
20+ engineered features
RSI, MACD histogram, MACD signal delta, Bollinger Band position, price vs SMA 20/50, EMA cross, volume ratio, momentum (1d/3d/5d), volatility, ATR, trend slope, and more.
Consensus mechanism
All four models vote on direction. When 3 or 4 agree, that consensus signal carries higher confidence. Disagreement is flagged as uncertain.
Minimum data requirement
Requires 61+ trading days of historical data per stock to train and run predictions. Stocks with insufficient history are skipped gracefully.
Runs alongside AI analysis
ML predictions run as a separate layer alongside AI analysis and technical indicators. The three sources can agree or diverge — both are shown to you.
From raw data to ML prediction
Five steps from historical OHLCV to a consensus directional call.
Historical data loaded
Up to 120 days of OHLCV data is loaded for the stock being analysed.
20+ features engineered
Technical features are computed from the raw data: momentum, volatility, indicator values, price ratios.
All 4 models run
Each model independently processes the feature set and produces a directional prediction.
Consensus score calculated
Agreement level across models is quantified. 4/4 agreement = very high confidence. 2/4 = uncertain.
Result added to your dashboard
ML prediction (direction + consensus score) is displayed alongside the AI signal and technical indicators.
ML predictions + advanced AI — better together
ML models find statistical patterns in historical price data. advanced AI reasons about current conditions, macro context, and fundamental factors. When both agree, conviction is high. When they diverge, you see both — so you can weigh the evidence yourself.
Related features
Four ML models running on your stocks.
Upgrade to Premium and unlock ML consensus predictions alongside AI analysis for every stock in your watchlist.