Back to Library

Perceptron Breakout Filter

AI / ML

A single-layer perceptron that learns in real-time which breakouts are likely to follow through vs. fakeout, using volume, squeeze, and momentum features.

How It Works

The indicator detects when price breaks above/below a Donchian channel (breakout). For each breakout, it computes a feature vector: relative volume at the breakout bar, Bollinger Band squeeze tightness, momentum (rate of change), and candle body ratio. A perceptron with online gradient descent predicts whether the breakout will follow through. After N bars, the actual outcome is measured and the perceptron weights are updated. Over time it learns which breakout conditions lead to real moves vs. traps on the specific instrument and timeframe.