Aurum AI

Aurum: Where Deep Learning Meets Gold

Aurum represents a new approach to gold trading. Rather than relying on rigid rule-based systems or lagging indicators, Aurum deploys a dual neural network architecture trained on over 59,000 hours of XAUUSD price action. The result is an expert advisor that learns patterns invisible to traditional technical analysis.

The Architecture

At its core, Aurum runs two specialized neural networks in parallel:

The first is a hybrid LSTM-CNN model. Long Short-Term Memory layers capture temporal dependencies across 30 bars of history, while Convolutional layers extract local patterns from the feature matrix. This combination allows Aurum to understand both the rhythm of price movement and the structure of market microstructure.

The second network is a profitability classifier. Before any trade is placed, this model evaluates the probability of a profitable outcome. Only when confidence exceeds the internal threshold does Aurum act. This selective approach means fewer trades, but higher quality setups.

What Aurum Sees

Every hour, Aurum processes 51 distinct features extracted from raw market data:

Price dynamics through Bollinger Bands, RSI, and MACD. Volatility regimes via Yang-Zhang, Parkinson, and Garman-Klass estimators. Market microstructure including tick volume patterns, order flow imbalance, and aggression ratios. Regime detection through Hurst exponent and approximate entropy calculations. Cross-asset correlations with silver. Temporal patterns encoded as cyclical features.

These features undergo correlation pruning and quality scoring before training, ensuring the model learns from meaningful signals rather than noise.

Training Methodology

Aurum was trained using gap validation, a technique that prevents data leakage by testing on a held-out middle section of the dataset. This ensures the model generalizes to unseen market conditions rather than memorizing historical patterns.

The training dataset spans from 2015 to 2026, covering multiple market regimes including the 2020 pandemic volatility and the subsequent gold rally to all-time highs above $4,500.

Performance Characteristics

Price prediction model metrics on test data:

• R-squared: 0.7845
• Mean Absolute Percentage Error: 5.15%
• Root Mean Square Error: 0.0141 (normalized)

Historical backtest results (2018-2025, H1 timeframe):

• Profit Factor: 1.58
• Maximum Drawdown: 8.35%
• Recovery Factor: 3.70

Risk Management

Aurum includes comprehensive risk controls. Position sizing is calculated as a percentage of account equity. Stop-loss and take-profit levels are determined dynamically using ATR-based triple barrier methodology. Daily loss limits and maximum drawdown thresholds provide additional protection. Session filters allow trading to be restricted to specific market hours.

Technical Requirements

• Platform: MetaTrader 5
• Symbol: XAUUSD
• Timeframe: H1
• Minimum Deposit: Recommended $1,000 or equivalent
• Broker: Any broker offering XAUUSD with reasonable spreads

What Aurum Is Not

Aurum is not a guarantee of profits. No trading system can predict the future with certainty. Market conditions change, and past performance does not ensure future results. Aurum is a tool that applies machine learning to identify potential opportunities. The decision to trade and the management of risk remains with the user.

Aurum does not use martingale, grid, or averaging strategies. Each trade stands on its own merit with defined risk parameters.

For Serious Traders

Aurum is designed for traders who understand that edge in the markets comes from disciplined execution of a well-researched strategy. The neural networks provide the analysis. The risk management provides the framework. The results depend on consistent application over time.

Before deploying on a live account, users should test Aurum on a demo account to understand its behavior and verify compatibility with their broker's execution.