This MetaTrader 5 expert advisor combines technical indicators with machine learning forecasts to trade the Forex market. It uses a trend-following approach based on Exponential Moving Average crossovers, filtered by the Average Directional Index and volume confirmations. A unique feature is its integration with extern
Trading robot · MT5
Cortex5 AI Trading
This artificial intelligence expert advisor for MetaTrader 5 utilizes a Double-Dueling Deep Recurrent Q-Network with LSTM memory to automate forex trading.
High risk
GitHub stars2
Forks0
Last update1 year ago20 Aug 2025
LicenceNone stated
Overview
The system processes a 45-dimensional state space to execute trades based on reinforcement learning. It includes extensive risk management features such as ATR-based dynamic stop losses, take profit targets, and a maximum holding time of 72 hours. To prevent overtrading, it employs a sophisticated cooldown system and choppy market detection. The EA requires a separate pre-trained model file to operate.
- Double-Dueling DRQN with LSTM recurrent neural network
- ATR-based dynamic stop loss and profit targets
- Multi-layer account protection and drawdown limits
- Advanced cooldown system with loss-based scaling
- Market regime and choppy price detection filters
Details
Author
toniibrahim on GitHub
Type
Trading robot
Platform
MetaTrader 5 (MQL5)
Strategy
AI / machine learning, Martingale
Markets
Forex
Timeframes
M15-H1, H4 and above
Risk profile
- Uses martingale lot scaling: losses can grow fast
EA files
Input parameters
Read from cortex5.mq5. Defaults are the author's; change them in the EA's Inputs tab.
| Name | Type | Default | Description |
|---|---|---|---|
InpModelFileName | string | DoubleDueling_DRQN_Model.dat | AI brain file (must be trained first) |
InpRiskPercent | double | 5.0 | Risk per trade as % of total account balance |
InpMaxSpread | double | 15.0 | Don't trade if broker spread exceeds this (points) |
InpATRMultiplier | double | 2.0 | Stop loss distance as multiple of Average True Range |
InpRRRatio | double | 1.5 | Risk:Reward ratio (1.5 = target 1.5x profit vs loss) |
InpMaxDrawdown | double | 10.0 | Pause trading if account loses more than this % |
InpMinWinRate | int | 30 | Pause if win rate drops below this % (over last 10 trades) |
InpCommissionPerLot | double | 0.0 | Your broker's commission per lot (for accurate cost calculation) |
InpUseRiskSizing | bool | true | Use smart position sizing based on volatility |
InpEnforceSymbolTF | bool | true | Only trade the symbol/timeframe the model was trained on |
InpAllowPositionScaling | bool | true | Allow adjusting position size based on signal strength |
InpMaxHoldingHours | int | 72 | Maximum hours to hold a position (Phase 1) |
Show all 40 inputs
| Name | Type | Default | Description |
|---|---|---|---|
InpProfitTargetATR | double | 2.5 | Take profit at N x ATR (Phase 1) |
InpUseProfitTargets | bool | true | Enable automatic profit taking (Phase 1) |
InpUseMaxHoldingTime | bool | true | Enable maximum holding time control (Phase 1) |
InpHoldingTimePenalty | double | 0.001 | Penalty per hour held in reward calculation (Phase 1) |
InpQuickExitBonus | double | 0.005 | Bonus for trades < 24 hours (Phase 1) |
InpFlatActionWeight | double | 1.5 | Increased weight for FLAT action in future training (Phase 2) |
InpEnhancedRewards | bool | true | Use enhanced reward calculation (Phase 2) |
InpDrawdownPenalty | double | 0.01 | Penalty for unrealized drawdown (Phase 2) |
InpUseDynamicStops | bool | true | Enable dynamic stop loss tightening (Phase 3) |
InpStopTightenRate | double | 0.8 | Stop tightening multiplier per day held (Phase 3) |
InpUsePositionFeatures | bool | true | Add position-aware features to state (Phase 3) |
InpUseMarketRegime | bool | true | Enable market regime detection (Phase 3) |
InpMinBarsBetweenTrades | int | 4 | Minimum bars between trades |
InpMaxTradesPerDay | int | 8 | Maximum trades per day |
InpEnforceFlat | bool | true | Force FLAT when position reaches limits |
InpUseAdvancedCooldown | bool | true | Enable enhanced cooldown system |
InpMinCooldownMinutes | int | 65 | Minimum cooldown period after closing position (minutes) |
InpUseChoppyDetection | bool | true | Enable choppy market detection |
InpChoppyATRThreshold | double | 0.8 | ATR ratio threshold for choppy markets (current vs 20-period avg) |
InpChoppyLookbackPeriod | int | 20 | Lookback period for choppy market analysis |
InpChoppyMinRange | int | 30 | Minimum points range to avoid micro-movements |
InpUseConsecutiveLosses | bool | true | Enable consecutive loss protection |
InpMaxConsecutiveLosses | int | 3 | Max consecutive losses before extended cooldown |
InpExtendedCooldownMin | int | 120 | Extended cooldown after consecutive losses (minutes) |
InpUseLossBasedCooldown | bool | true | Scale cooldown based on loss size |
InpUseSignalRefinement | bool | true | Enable enhanced signal processing |
InpUseSignalSmoothing | bool | true | Enable signal smoothing over multiple bars |
InpSignalSmoothingPeriod | int | 3 | Number of bars for signal averaging |
How to install it
- 1Download the sourceUse "Download source (.zip)" above, or clone toniibrahim/cortex5-ai-trading from GitHub, and unzip it.
- 2Open the data folderIn MetaTrader choose File, then Open Data Folder, and go to MQL5/Experts.
- 3Copy the filesCopy the .mq5 file into Experts, and any .mqh files into the Include folder the code expects.
- 4Compile in MetaEditorOpen the file in MetaEditor and press Compile. Fix any missing includes the compiler reports.
- 5Test before you tradeRun it in the Strategy Tester, then attach it to a demo chart and allow algorithmic trading.
Third-party code. ForexR did not write, test or endorse this EA, and the summary above was generated automatically from its README and code. ForexR lists only repositories with readable source code and never hosts files. Read the code, backtest it and use a demo account first; automated trading can lose money quickly.
Similar expert advisors
This machine learning robot for MetaTrader 5 uses an ensemble of AI models including LSTM, GRU, and LightGBM to predict price movements. It executes trades based on signal confidence and predicted price targets. The EA features multiple risk controls such as volatility filters, trend-based MA filters, RSI checks, and a
This MetaTrader 5 expert advisor utilizes an Echo State Network, which is a type of recurrent neural network, to analyze price action and execute trades. The system computes various statistical inputs, including price volatility, mean values, and Hurst exponent metrics, to feed its reservoir computing model. Risk is ma
This MetaTrader 5 expert advisor is a legacy project focused on machine learning and neural network forecasting. The system implements logic for training models and generating forecasts using technical indicators like ADX and ZigZag. It includes several optional modules for long, short, hedging, and martingale trading
LQS AI Institutional Engine is a MetaTrader 5 robot designed for trading gold using market microstructure analysis. It combines four analytical modules focusing on liquidity sweeps, order flow divergence, Wyckoff accumulation and distribution phases, and multiple anchored VWAP levels. The system features an integrated
This machine learning expert advisor for MetaTrader 5 utilizes ensemble and deep learning models to analyze market patterns. It offers a versatile multi-strategy approach, allowing users to toggle between trend-following, scalping, and grid-hedging behaviors. Risk controls are comprehensive, featuring fixed percentage
Need a broker for MT5?
Compare regulated brokers that support MetaTrader, with spreads, minimum deposits and leverage side by side.
Compare brokers