How the System Works

Triple-Engine Architecture

Our forex intelligence system combines three distinct AI approaches to generate high-confidence trading predictions:

• Traditional Technical Analysis Engine: Traditional technical analysis using EMA, MACD, RSI, Bollinger Bands, and ATR indicators. Expected accuracy: 70-75%. Rule-based with proven track record.
• ML XGBoost Engine: Machine learning using gradient boosting algorithm. Trains on verified predictions to recognise patterns. Expected accuracy: 80-85% after training on 100+ verified predictions.
• GARCH-LSTM Engine: Advanced neural network combining GARCH volatility forecasting with LSTM deep learning. Includes Bayesian confidence intervals. Expected accuracy: 85-92%.

Signal Confluence

When 2 or more engines agree on the same signal (BUY, SELL, or HOLD), we create a "consensus signal" with weighted confidence. Historical data shows:

• 2/3 engines agree: 85-88% accuracy
• 3/3 engines agree: 90-95% accuracy

Data Collection

The system fetches live forex rates from Chainlink oracles every 5 minutes for 8 major pairs:

• Major pairs: EUR/USD, GBP/USD, USD/JPY
• Commodity currencies: AUD/USD, NZD/USD, USD/CAD, USD/CHF
• Cross pair: EUR/GBP

Prediction Cycle

Every 30 minutes, all three engines analyse the latest data:

1. Traditional Techncial Analysis engine generates technical analysis-based predictions
2. ML engine uses those indicators plus its trained model
3. GARCH engine analyses price history for volatility and trends
4. Predictions are compared and consensus is calculated
5. After 4 hours, predictions are verified against actual price movement

Accuracy Tracking

Every prediction is verified after 4 hours. A prediction is marked "correct" if:

• BUY signal: Price moved up by 0.1% or more
• SELL signal: Price moved down by 0.1% or more
• HOLD signal: Price stayed within ±0.1% range

Our 83.4% system accuracy is calculated from 17,163 correct predictions out of 20,579 total verifications. We don't hide failures.

Limitations & Responsible Use

• Prediction ≠ Certainty: Even 90% accuracy means 10% of predictions fail. Forex is volatile.
• Black Swan Events: No model predicts major geopolitical events, central bank surprises, or flash crashes.
• Data Dependency: Engines require 50+ data points (4-6 hours) to start generating predictions.
• Training Period: ML and GARCH engines improve over time as they collect verified predictions.
• Not Financial Advice: This dashboard provides data-driven insights, not investment recommendations.

Roadmap

Upcoming enhancements:

• Multi-Timeframe Analysis: 15-min, 1-hour, 4-hour, and daily predictions
• Risk Management Tools: Automated stop-loss and take-profit calculations
• Alert System: Email/Telegram notifications for high-confidence consensus predictions
• Backtesting Interface: See how predictions performed historically
• Portfolio Optimisation: Recommend position sizing based on confidence levels