Artificial Intelligence
This EA uses a neural network-inspired approach to predict price movement directions, employing a multi-layer perceptron (MLP) architecture trained on historical price data. It generates buy/sell signals based on the network's output after processing normalized input features derived from price and indicator values. The system autonomously opens and manages positions according to its trained model predictions.
Open the interactive view (equity curve, symbol switch)Key facts
Backtest results (H1, about 1 year)
| Symbol | P&L | Annual | Max DD | Trades | Win rate | Sharpe | Recovery | Losing streak | Per trade | Status |
|---|---|---|---|---|---|---|---|---|---|---|
| EURUSD | -5.66% | -4.94% | 8.28% | 96 | 43.20% | -1.24 | -0.68 | 9 | −$5.60 | Reliable |
| GBPUSD | -3.57% | -3.11% | 5.63% | 97 | 48.50% | -0.64 | -0.63 | 7 | −$3.68 | Reliable |
| USDJPY | +5.94% | +5.14% | 3.14% | 100 | 48.50% | 1.18 | 1.89 | 6 | +$5.84 | Reliable |
| XAUUSD | -54.96% | -48.27% | 83.47% | 89 | 47.70% | 0.45 | -0.66 | 8 | −$59.64 | Reliable |
Profitable on 1 of 4 markets with reliable results.
Converted from MQL5 to Python and run with the same rules for every EA. Positions use MT5 lot sizes (100,000 units per FX lot, 100 oz per gold lot) with 1:100 leverage, one order at a time, and a stop-out below 10% of the starting balance. Sharpe is computed from the equity curve. Results marked as unreliable are shown for completeness only.
Month by month
| Market | Months with a gain | Best month | Worst month |
|---|---|---|---|
| EURUSD | 31% | +1.27% 2026-02 | -1.86% 2026-10 |
| GBPUSD | 46% | +1.76% 2026-08 | -2.80% 2026-09 |
| USDJPY | 54% | +2.46% 2026-04 | -2.50% 2026-08 |
| XAUUSD | 38% | +105.92% 2026-02 | -35.97% 2025-12 |
From the backtest equity curve, sampled about every two days; only reliable results.
Blow-up risk: No blow-up pattern found
None of the usual blow-up patterns were found in the code, and the account was not stopped out in our backtests. That is not a guarantee: any EA can lose money.
High: stopped out in our backtest, or martingale/grid sizing without a stop loss. Elevated: martingale or grid sizing with a stop loss, or several positions at once without a stop loss. What the label means and what we found →
Strategy analysis
The strategy concept is based on a multi-layer perceptron (MLP) artificial neural network model that receives normalized price features and indicator data through the input layer, undergoes nonlinear transformations in hidden layers, and produces trading direction predictions at the output layer. Entry conditions trigger buy or sell signals when the neural network output exceeds predefined threshold values, typically combining multiple technical indicators such as moving averages, RSI, and MACD as input features. Exit conditions rely on take-profit and stop-loss settings or the generation of opposite signals, with some versions also incorporating trailing stop functionality. Position sizing employs fixed lot sizes or account balance-proportional lot calculations, but generally lacks advanced money management strategies such as the Kelly criterion or dynamic risk adjustment. In terms of money management, the EA's risk control primarily depends on static stop-loss and lot size settings, lacking an adaptive position adjustment mechanism to respond to changes in market volatility, with overall risk exposure depending on the network's prediction accuracy and stop-loss distance configuration.
Strengths
- Leverages machine learning methods to adaptively learn market patterns, theoretically capable of capturing nonlinear price relationships
- Multi-indicator input features provide a multi-dimensional market perspective, enhancing signal reliability
- Neural network can be periodically retrained to adapt to changing market conditions
- Highly automated, completing the full process from analysis to trading without manual intervention
- Built-in stop-loss and take-profit mechanisms provide basic risk protection
Weaknesses
- Neural networks are prone to overfitting historical data, and live performance may fall far short of backtesting results
- Lacks advanced money management strategies; fixed lot sizing cannot dynamically adjust risk exposure based on market volatility
- Training data volume and quality significantly impact model effectiveness; insufficient data leads to notably degraded prediction ability
- Black-box nature makes strategy logic difficult to explain and debug; hard to identify root causes when losses occur
- Does not include adequate slippage and spread filtering, which may cause actual execution to deviate from expectations in high-frequency trading environments
Possible improvements
- Add cross-validation and out-of-sample testing mechanisms to prevent overfitting; introduce regularization techniques such as Dropout or L2 regularization
- Implement dynamic position sizing that adaptively adjusts lot size based on current account equity, market volatility (ATR), and model confidence
- Add a market regime identification module to switch between different model parameters in different market environments (trending/ranging/high volatility)
- Add slippage protection and spread filtering mechanisms to prohibit opening positions when liquidity is insufficient or spread is too wide
Risk assessment
Overall risk: Medium
- Neural network overfitting risk leads to unstable live performance; historical backtesting may significantly overestimate strategy returns
- Lack of dynamic stop-loss adjustment mechanism; fixed stop-loss may be insufficient to protect the account duringsevere volatile market conditions
- Unexplainability of black-box models increases difficulty of quickly identifying and fixing issues when the strategy fails
- No filtering for trading sessions and market liquidity; may generate false signals during low-liquidity periods
Evaluation scores
Parameters
| Parameter | Default | Description |
|---|---|---|
Lots | 0.1 | Fixed lot size per trade |
StopLoss | 50 | Stop loss in points |
TakeProfit | 100 | Take profit in points |
TrailingStop | 0 | Trailing stop in points, 0 to disable |
InputLayerSize | 8 | Neural network input layer nodes |
HiddenLayerSize | 6 | Neural network hidden layer nodes |
LearningRate | 0.01 | Neural network learning rate |
Threshold | 0.5 | Signal trigger threshold |
TrainingPeriod | 1000 | Training history period (bars) |
Similar Machine Learning EAs
Backtests are historical simulations and do not guarantee future results. Not financial advice.