AI Features

Overview

Every signal passes through three AI checks before an order is placed:

1. 1

   ML Signal Filter

   Gradient-boosted classifier trained on your own trade history

2. 2

   Sentiment Analysis

   News headlines scored — contradicting signals blocked

3. 3

   Volatility Prediction

   GARCH + ATR percentile — skips low-vol, adjusts lot size in high-vol

1 — ML Signal Filter

A Gradient Boosted classifier trained on your closed trade history. It learns which market conditions and signal features actually produced wins on your specific account and broker — and blocks setups that historically lose.

Features the model uses:

Signal score (0–100)     R:R ratio
Session (London/NY/overlap)   ADX at signal time
RSI at signal time       ATR ratio (normalised volatility)
Strategy type            Direction (buy/sell)
Hour of day (UTC)

How it trains:

Every closed trade → outcome recorded to ai/model_cache/trade_history.csv
After 30 trades → first model trained automatically
Every 7 days    → model retrained on latest history
At inference    → P(win) ≥ 55% required to allow trade

2 — Sentiment Analysis

Pulls recent forex headlines from free RSS feeds every 30 minutes and scores them as bullish or bearish using keyword matching. Strong news that contradicts the trade direction is blocked.

Sources: ForexFactory RSS, FXStreet RSS, Yahoo Finance RSS
         FinnHub API (optional — add token in settings.json)

Blocking rules:
  Buy signal  + sentiment score ≤ −0.4  →  blocked (bearish news)
  Sell signal + sentiment score ≥ +0.4  →  blocked (bullish news)

Neutral range (−0.4 to +0.4) → trade allowed

To enable better news coverage, add a free FinnHub token:

{
  "notifications": {
    "finnhub_token": "your_free_token_from_finnhub.io"
  }
}

3 — Volatility Prediction

Predicts next-bar volatility using two methods combined: ATR percentile ranking (where is current ATR relative to the last 90 bars?) and a GARCH(1,1) one-step-ahead variance forecast.