🤖AI & Machine Learning

Neural Network Signal Confidence Scoring

BananaEA v4.4.6 introduces an advanced Neural Network trained on 3,777+ historical DAX signals to score every trade signal with a confidence percentage. This allows you to filter out low-quality signals and only trade when the AI has high confidence.

---

🎯 MinimumAIConfidence Parameter

What It Does

Every signal detected by BananaEA (Bu1, Bu2, Be1, Be2, etc.) is evaluated by an 11-feature Neural Network that assigns a confidence score from 0-100%. The MinimumAIConfidence parameter controls which signals are allowed to become trades.

Configuration

MinimumAIConfidence = 0      // Default: Take ALL signals (AI disabled)
MinimumAIConfidence = 50     // Moderate: Filter bottom 50% of signals
MinimumAIConfidence = 65     // Aggressive: Take only high-confidence signals
MinimumAIConfidence = 80     // Very conservative: Top 20% signals only

Default Value

• Setting: 0 (AI filtering disabled)

• Why: Allows users to experience EA behavior without AI filtering first

• Recommendation: Start with 0, then gradually increase to 50 → 65 → 80 based on results

---

🧠 How Neural Network Scoring Works

11 Neural Network Features

The AI analyzes these factors for every signal:

1. Signal Strength: Calculated momentum score (1.0-2.1 scale)

2. Volatility Classification: Low/Medium/High/Extreme market conditions

3. Trading Session Context: Frankfurt Open, European Morning, US Session, etc.

4. European Session Score: DAX-specific weighting (0.3-1.0)

5. Trend Following Indicators: Momentum alignment (0.4-0.7)

6. Mean Reversion Signals: Reversion probability (0.4-0.8)

7. Institutional Bias Detection: Smart money positioning (0.6-0.9)

8. Volatility Preference: Optimal volatility conditions (0.5-0.8)

9. Timeframe Context: Current timeframe characteristics

10. Cross-Timeframe Analysis: Multi-timeframe correlation

11. Historical Pattern Matching: Similar past signal outcomes

Confidence Calculation Process

1. Signal Detected → Bu1, Be2, or other pattern found

2. Feature Extraction → 11 features calculated from market data

3. Neural Network Scoring → Model outputs 0-100% confidence

4. Threshold Check → If confidence ≥ MinimumAIConfidence, trade proceeds

5. Trade Execution → Only high-confidence signals become trades

---

📊 Performance Impact: Real Backtest Results

Without AI Filtering (MinimumAIConfidence = 0)

• Win Rate: 48.3%

• Total Trades: 1,247

• Profit Factor: 1.82

With AI Filtering (MinimumAIConfidence = 65)

• Win Rate: 54.8% ✅ +6.5% improvement

• Total Trades: 743 (59.5% of original)

• Profit Factor: 2.34 ✅ +28.6% improvement

Key Insight: AI filtering removes 40% of trades but those trades were predominantly losers, resulting in significantly higher win rates and profit factors.

---

⚙️ Configuration Strategies

Strategy 1: Aggressive Trading (Default)

MinimumAIConfidence = 0

Use When:

• Learning EA behavior and signal patterns

• Maximum trade frequency desired

• Testing new symbols or timeframes

• Building signal database for analysis

Pros: Maximum opportunities, full signal visibility Cons: Lower win rate, more false signals

---

Strategy 2: Balanced Filtering

MinimumAIConfidence = 50

Use When:

• Standard live trading with prop firms

• Seeking balance between frequency and quality

• Moderate risk tolerance

• Building consistent equity curve

Pros: Good trade frequency, improved quality Cons: Misses some marginal winners

---

Strategy 3: High-Confidence Only

MinimumAIConfidence = 65

Use When:

• Passing prop firm challenges (need high win rate)

• Conservative account management

• Trading during volatile/uncertain markets

• Focus on quality over quantity

Pros: Highest win rate, best profit factor Cons: Fewer trading opportunities

---

Strategy 4: Ultra-Conservative

MinimumAIConfidence = 80

Use When:

• Extreme risk aversion

• Prop firm final evaluation stage

• Limited trading capital

• Building psychological confidence

Pros: Maximum signal quality, minimal losing trades Cons: Very low trade frequency, may miss trends

---

🔍 Signal Quality Distribution

Based on 3,777 analyzed DAX M5 signals:

Confidence Range	% of Signals	Win Rate	Recommendation
80-100%	18%	62.3%	✅ Excellent - Trade all
65-79%	23%	56.1%	✅ Good - Trade most
50-64%	31%	51.2%	⚠️ Fair - Selective
0-49%	28%	39.7%	❌ Poor - Avoid

Interpretation: Signals below 50% confidence have negative edge (< 50% win rate). Filtering these out dramatically improves performance.

---

🎯 Optimization Strategy

Step 1: Establish Baseline

1. Run 3-month backtest with MinimumAIConfidence = 0

2. Record win rate, profit factor, total trades

3. Analyze which signals won/lost most frequently

Step 2: Apply Moderate Filtering

1. Set MinimumAIConfidence = 50

2. Re-run same backtest period

3. Compare metrics: Did win rate improve? By how much?

Step 3: Test Aggressive Filtering

1. Set MinimumAIConfidence = 65

2. Backtest again

3. Verify trade frequency is still acceptable (minimum 50-100 trades in 3 months)

Step 4: Live Testing

1. Start with conservative setting (MinimumAIConfidence = 65)

2. Monitor first 20 live trades

3. Adjust based on results: Lower if too few trades, raise if too many losers

---

💡 Best Practices

✅ DO:

• Start conservative (65+) and lower if needed

• Backtest thoroughly before live trading with AI filtering

• Track confidence scores of winning vs. losing trades

• Adjust seasonally - volatile markets may need higher thresholds

• Use with other filters - AI + ATR + Daily Range = powerful combination

❌ DON'T:

• Set too high initially (80+) - may miss too many opportunities

• Change during drawdown - emotional adjustments often backfire

• Ignore trade frequency - Need minimum 50 trades per quarter for statistical validity

• Use without backtesting - Always validate AI filtering on historical data first

• Disable on losing streaks - AI filtering is designed for long-term edge, not short-term fixes

---

🔧 Troubleshooting

Problem: No Trades Generated

Cause: MinimumAIConfidence set too high (e.g., 90-100)

Solution:

1. Lower to 65 and verify trades appear

2. Check EA logs for "Signal rejected: AI confidence below threshold" messages

3. Ensure Neural Network model is loaded (check initialization logs)

---

Problem: Still Getting Too Many Losing Trades

Cause: Other filters may be disabled or AI threshold too low

Solution:

1. Increase MinimumAIConfidence to 70-75

2. Enable complementary filters:

   • UseATRFilter = true (volatility filter)

   • UseDailyRangeFilter = true (range filter)

   • Use10OClockFilter = true (directional bias filter)

3. Review optimization set - may need updated parameters

---

Problem: Confidence Scores All Show 0%

Cause: Neural Network model not loading or Python dependencies missing

Solution:

1. Model is built-in - no Python needed for scoring

2. Check MT4 Experts log for error messages

3. Verify EA version is v4.4.6 or higher

4. Reinstall EA if necessary (model embedded in .ex4 file)

---

📈 Advanced: Combining AI with Other Filters

Conservative Day Trading Setup

// AI Filtering
MinimumAIConfidence = 65

// Volatility Control
UseATRFilter = true
MaxATRMultiplier_Banana = 2.0

// Range Management
UseDailyRangeFilter = true
MaxDailyRangePercent = 100.0

// Session Control
UseSessionMovementFilter = true
MaxSessionMovementPercent = 20.0

Result: Only highest-quality signals during optimal market conditions

---

Aggressive Scalping Setup

// AI Filtering
MinimumAIConfidence = 50

// Volatility Control
UseATRFilter = false

// Range Management
UseDailyRangeFilter = false

// Spread Control (critical for scalping)
MaxSpread = 2.0

Result: Higher trade frequency with moderate quality filtering

---

🎓 Understanding the Training Data

Data Source

• Signals: 3,777 DAX M5 signals from 2023-2024

• Patterns: Bu1, Bu2, Bu3, Be1, Be2, Be3, TP, TTP, TTTP

• Timeframes: M5 primary, M15/H1 correlation

• Sessions: All trading hours (Frankfurt, London, New York)

Model Validation

• Training Set: 80% (3,021 signals)

• Validation Set: 20% (756 signals)

• Cross-Validation: 5-fold validation for robustness

• Accuracy: 78.4% on validation set

Model Updates

• Frequency: Quarterly retraining with latest market data

• Distribution: Automatic with EA updates

• Backwards Compatible: Old models work but may be less accurate

---

🔗 Related Settings

• Signal Configuration - Enable/disable specific patterns

• Advanced Features - Combine with BE/PC/TS for optimal risk management

• Trading Hours & Filters - ATR and Daily Range filters complement AI scoring

• Risk Management - Adjust position sizing based on signal confidence

---

📚 Further Reading

• AI Optimization Guide - Python AI integration for parameter discovery

• AI Signal Scoring Guide - Deep dive into Neural Network architecture

• Professional Metrics Guide - Sharpe/Calmar/Recovery Factor optimization

• FAQ: AI & Machine Learning - Common AI-related questions