Autoencoder Anomaly Detection for Market Manipulation

A browser-based JavaScript application that uses MLP (Multi-Layer Perceptron) autoencoders to detect anomalous market conditions in cryptocurrency trading data from Binance API.

Overview

This application implements an unsupervised learning approach to identify market manipulation patterns such as:

• Liquidation cascades
• Unusual funding spikes
• Abnormal volume patterns
• Price manipulation attempts

The system uses a neural autoencoder trained on "normal" market periods and flags periods with high reconstruction error as potential anomalies.

Features

Core Functionality

• Real-time Data Fetching: Direct integration with Binance API for live market data
• Batch Data Processing: Efficient handling of large date ranges with automatic batching
• MLP Autoencoder: Deep neural network architecture for pattern recognition
• Ensemble Autoencoder: Multiple model voting system for improved accuracy
• Advanced Feature Engineering: 12 comprehensive market features including RSI, Bollinger Bands, and market regime detection

Data Management

• Flexible Date Ranges: Custom date range selection or default lookback periods
• Multiple Time Intervals: Support for 1h, 4h, 1d, and 1w intervals
• Data Quality Analysis: Automatic detection of missing data, zero volume, and corrupted records
• Raw Data Export: Download processed data as JSON files
• MinMax Normalization: Consistent data scaling for optimal training

Anomaly Detection

• Multi-level Severity Classification: Normal, Warning, and Critical anomaly levels
• Sensitive Threshold Detection: Optimized thresholds for catching major market events
• Feature Contribution Analysis: Detailed breakdown of which market features contribute to anomalies
• Major Event Detection: Specialized algorithms for identifying crash patterns and market manipulation
• Historical Event Analysis: Automatic detection of significant market events

Visualization & Analysis

• Interactive Charts: Training progress, reconstruction errors, and severity distribution
• Price & Volume Timeline: Dual-axis charts showing price and volume trends
• Anomaly Timeline: Color-coded scatter plots showing anomaly distribution over time
• Severity Distribution: Doughnut charts showing anomaly breakdown
• Feature Contribution Charts: Visual analysis of feature importance

User Interface

• Modern Dark Theme: Professional gradient-based UI with glassmorphism effects
• Responsive Design: Mobile-friendly interface with adaptive layouts
• Real-time Progress Tracking: Live progress bars and status updates
• Interactive Controls: Customizable training parameters and data range selection
• Data Sample Tables: Preview of raw and processed data
• Statistics Dashboard: Comprehensive data quality and performance metrics

Model Management

• Model Persistence: Save and load trained models using IndexedDB
• Configuration Storage: Persistent storage of model parameters and thresholds
• Training History: Track and visualize training progress over epochs
• Early Stopping: Automatic training termination to prevent overfitting
• Memory Management: Efficient tensor disposal and memory cleanup

Feature Selection & Engineering

Current Encoder Features (12 Features)

The autoencoder encoder currently uses 12 carefully selected market features optimized for anomaly detection:

Core Price Features

1. priceChange - Close-to-close percentage change

   • Formula: ((close - previous_close) / previous_close) * 100
   • Purpose: Captures basic price movements and trends

2. priceAcceleration - Second derivative of price changes

   • Formula: priceChange - previous_priceChange
   • Purpose: Detects sudden price accelerations (crash patterns)

3. priceGap - Gap between opening price and previous close

   • Formula: |open - previous_close| / previous_close * 100
   • Purpose: Identifies overnight gaps and flash crashes

4. priceMomentum - 3-period price momentum

   • Formula: (close - close_3_periods_ago) / close_3_periods_ago * 100
   • Purpose: Captures medium-term price trends

Volume Features

5. volume - Raw trading volume (MinMax normalized)

   • Purpose: Market activity and liquidity indicator

6. volumeSpike - Volume spikes above 30% threshold

   • Formula: max(0, volumeChange - 30)
   • Purpose: Detects unusual volume surges

7. volumeMomentum - 3-period volume momentum

   • Formula: (volume - volume_3_periods_ago) / volume_3_periods_ago * 100
   • Purpose: Captures volume trend changes

Technical Indicators

8. rsi - 14-period Relative Strength Index

   • Range: 0-100 (normalized to 0-1)
   • Purpose: Momentum oscillator for overbought/oversold conditions

9. bollingerPosition - Position within Bollinger Bands

   • Range: -1 (lower band) to 1 (upper band)
   • Purpose: Volatility and mean reversion indicator

10. marketRegime - Market trend detection

    • Values: -1 (downtrend), 0 (ranging), 1 (uptrend)
    • Purpose: Contextual market state classification

Simulated Features

11. fundingRateProxy - Simulated funding rate for spot markets

    • Formula: (momentumFactor * 0.5 + volumeFactor * 0.3 + volatilityFactor * 0.2) * 0.01
    • Purpose: Approximates futures funding rate behavior

12. openInterestProxy - Simulated open interest

    • Formula: (volumeFactor * 0.4 + tradeIntensity * 0.3 + priceMomentum * 0.3)
    • Purpose: Estimates market positioning

Feature Selection Recommendations

Core Feature Set (8 Features) - Recommended

For optimal performance with reduced complexity:

const coreFeatures = [
    'priceChange',      // Essential price movement
    'volume',           // Market activity
    'priceAcceleration', // Sudden price changes
    'volumeSpike',     // Unusual volume
    'priceGap',        // Price gaps
    'rsi',            // Momentum oscillator
    'bollingerPosition', // Volatility position
    'marketRegime'    // Trend context
];

Enhanced Feature Set (10 Features)

For comprehensive analysis with moderate complexity:

const enhancedFeatures = [
    'priceChange',      // Price movement
    'volume',           // Volume
    'priceAcceleration', // Price acceleration
    'volumeSpike',     // Volume spikes
    'priceGap',        // Price gaps
    'priceMomentum',   // Price momentum
    'volumeMomentum',  // Volume momentum
    'rsi',            // RSI
    'bollingerPosition', // Bollinger position
    'marketRegime'    // Market regime
];

Feature Importance Analysis

High Importance Features ⭐⭐⭐

• priceChange - Fundamental price movement indicator
• priceAcceleration - Critical for crash detection
• volumeSpike - Direct anomaly indicator
• priceGap - Flash crash detection

Medium Importance Features ⭐⭐

• volume - Market activity baseline
• rsi - Momentum context
• bollingerPosition - Volatility context
• marketRegime - Trend context

Lower Importance Features ⭐

• priceMomentum - Redundant with priceAcceleration
• volumeMomentum - Redundant with volumeSpike
• fundingRateProxy - Limited value for spot markets
• openInterestProxy - Simulated feature

Architecture Optimization

When reducing features, adjust the autoencoder architecture:

// For 8 features
this.inputSize = 8;
this.encoderUnits = [12, 6, 3];
this.latentSize = 2;
this.decoderUnits = [3, 6, 12];

// For 10 features
this.inputSize = 10;
this.encoderUnits = [14, 7, 3];
this.latentSize = 2;
this.decoderUnits = [3, 7, 14];

Alternative Features to Consider

Volatility Features

• volatility - Intraday price range: ((high - low) / close) * 100
• volatilityRatio - Current vs average volatility

Volume Features

• volumeRatio - Current vs average volume
• buySellRatio - Taker buy vs sell volume ratio

Price Features

• vwapDeviation - Deviation from Volume Weighted Average Price
• priceRangePosition - Position within daily range: (close - low) / (high - low)

Feature Selection Testing

Ablation Studies

1. Train models with different feature combinations
2. Compare reconstruction errors and anomaly detection accuracy
3. Analyze feature contribution scores for each anomaly

Cross-Validation

1. Test on different time periods (bull/bear markets)
2. Validate on different trading pairs
3. Measure performance across various market conditions

Performance Metrics

• Reconstruction Error: Lower is better
• Anomaly Detection Rate: Balance between sensitivity and specificity
• False Positive Rate: Minimize false alarms
• Training Time: Faster training with fewer features

Feature Selection Configuration

A comprehensive feature selection configuration file (feature-selection-config.js) is provided with:

• Predefined Feature Sets: Core, Enhanced, Minimal, and Alternative feature combinations
• Architecture Configurations: Optimized network architectures for different feature counts
• Feature Descriptions: Detailed formulas and importance rankings
• Testing Utilities: Ablation studies and cross-validation configurations

Usage Example:

import { CORE_FEATURES, ARCHITECTURE_CONFIGS } from './feature-selection-config.js';

// Use core features (8 features)
const config = ARCHITECTURE_CONFIGS[CORE_FEATURES.length];
this.inputSize = config.inputSize;
this.encoderUnits = config.encoderUnits;

Exploratory Data Analysis (EDA) Summary

Data Overview

The application performs comprehensive EDA on cryptocurrency market data to understand patterns, detect anomalies, and ensure data quality before model training.

Data Collection Metrics

• Default Period: 60 days of historical data
• Time Intervals: 1h, 4h, 1d, 1w (configurable)
• Data Points: ~1,440 hourly records (60 days × 24 hours)
• Trading Pairs: BTC/USDT, ETH/USDT, BNB/USDT, ADA/USDT, SOL/USDT
• API Source: Binance REST API with batch processing support

Data Quality Assessment

The system automatically evaluates data quality with the following metrics:

• Data Completeness: Percentage of non-missing records
• Zero Volume Detection: Count of periods with zero trading volume
• Negative Price Detection: Count of invalid price records
• Missing Data Points: Count of incomplete OHLCV records
• Outlier Detection: Z-score threshold of 3.0 for extreme values

Statistical Analysis

Price Statistics

• Mean Price: Average closing price over the period
• Standard Deviation: Price volatility measure
• Min/Max Prices: Price range boundaries
• Price Change Distribution: Statistical analysis of price movements

Volume Statistics

• Mean Volume: Average trading volume
• Volume Volatility: Standard deviation of volume changes
• Volume Range: Min/max volume values
• Volume Change Distribution: Analysis of volume fluctuations

Change Statistics

• Price Change Mean: Average percentage price change
• Price Change Std: Volatility of price changes
• Volume Change Mean: Average percentage volume change
• Volume Change Std: Volatility of volume changes

Feature Engineering Analysis

Technical Indicators

• RSI Analysis: 14-period Relative Strength Index distribution
• Bollinger Bands: Position analysis within volatility bands
• Market Regime: Trend detection (uptrend/ranging/downtrend)

Derived Features

• Price Acceleration: Second derivative analysis for crash detection
• Volume Spikes: Detection of unusual volume increases (>30%)
• Price Gaps: Analysis of overnight price gaps
• Momentum Indicators: 3-period price and volume momentum

Historical Event Detection

Major Market Events

The system automatically identifies significant market events:

• Major Price Drops: >5% price decreases (Warning: -5% to -10%, Critical: <-10%)
• Major Volume Spikes: >100% volume increases (Warning: 100-200%, Critical: >200%)
• Crash Patterns: Price drop + volume spike combinations
• Extreme Events: Sensitive detection for major market disruptions

Event Classification

• Critical Events: High-impact market disruptions
• Warning Events: Moderate market anomalies
• Normal Events: Regular market behavior

Data Visualization

Price & Volume Timeline

• Dual-axis Charts: Price (left) and normalized volume (right)
• Interactive Tooltips: Detailed point information
• Trend Analysis: Visual identification of market trends

Statistical Distributions

• Histogram Analysis: Distribution of price and volume changes
• Box Plot Visualization: Quartile analysis and outlier detection
• Correlation Analysis: Relationships between different features

Data Preprocessing Pipeline

Normalization Process

• MinMax Scaling: All features normalized to [0,1] range
• Feature Scaling: Consistent scaling across all 12 features
• Outlier Handling: Robust scaling to handle extreme values

Train/Validation Split

• Training Data: 80% of chronological data
• Validation Data: 20% of chronological data
• Temporal Split: Maintains chronological order for time series

Anomaly Detection EDA

Reconstruction Error Analysis

• Error Distribution: Statistical analysis of reconstruction errors
• Threshold Calculation: Mean + 1.5 × Standard Deviation
• Severity Classification: Normal/Warning/Critical levels

Feature Contribution Analysis

• Individual Contributions: Which features drive anomalies
• Pattern Recognition: Common anomaly patterns
• Root Cause Analysis: Understanding anomaly sources

Data Quality Metrics

Completeness Metrics

• Missing Data Rate: Percentage of incomplete records
• Data Integrity: Validation of OHLCV consistency
• Temporal Continuity: Gap detection in time series

Quality Indicators

• Green (Good): >95% completeness, no data issues
• Yellow (Warning): 90-95% completeness, minor issues
• Red (Critical): <90% completeness, significant issues

Market Regime Analysis

Trend Detection

• Uptrend: Strong positive price momentum with low volatility
• Downtrend: Strong negative price momentum with low volatility
• Ranging: High volatility with minimal directional movement

Volatility Analysis

• Low Volatility: Stable price movements
• High Volatility: Erratic price movements
• Volatility Clustering: Periods of sustained high/low volatility

Performance Metrics

Data Processing Performance

• Fetch Time: 2-5 seconds (single request), 10-30 seconds (batch)
• Processing Time: <1 second for feature engineering
• Memory Usage: ~50-100MB during processing

Analysis Accuracy

• Event Detection Rate: Sensitivity to major market events
• False Positive Rate: Minimization of false alarms
• Pattern Recognition: Accuracy in identifying market patterns

EDA Output Examples

Sample Statistics Output

Price Statistics:
- Mean: $45,230.50 USDT
- Std Dev: $2,150.30 USDT
- Min: $38,900.00 USDT
- Max: $52,100.00 USDT

Volume Statistics:
- Mean: 1,250,000 units
- Std Dev: 450,000 units
- Min: 200,000 units
- Max: 3,500,000 units

Data Quality Report

Data Quality Assessment:
- Completeness: 98.5%
- Missing Data: 22 records
- Zero Volume: 0 records
- Negative Prices: 0 records
- Status: ✅ Excellent Quality

This comprehensive EDA framework ensures data quality, identifies patterns, and provides insights essential for effective anomaly detection in cryptocurrency markets.

Architecture

Data Layer (binance-data-loader.js)

• Batch Data Fetching: Handles large date ranges with automatic API batching (up to 1000 records per request)
• Multiple Time Intervals: Support for 1s, 1m, 3m, 5m, 15m, 30m, 1h, 2h, 4h, 6h, 8h, 12h, 1d, 3d, 1w, 1M intervals
• Advanced Feature Engineering: Calculates 12 comprehensive features:
  • Price change percentage and acceleration
  • Volume changes and spikes
  • Funding rate proxy (simulated for spot markets)
  • Open interest proxy (simulated for spot markets)
  • Price gaps and momentum indicators
  • Volume momentum analysis
  • RSI (Relative Strength Index) calculation
  • Bollinger Bands position
  • Market regime detection (trending vs ranging)
• Data Quality Control: Automatic detection and reporting of data quality issues
• MinMax Normalization: Consistent scaling across all features for optimal training
• Historical Event Analysis: Built-in detection of major market events and crash patterns

Model Layer (autoencoder.js)

• MLP Autoencoder: Multi-layer perceptron with encoder-decoder architecture
  • Encoder: 12 → 16 → 8 → 4 → 2 neurons (compression to latent space)
  • Decoder: 2 → 4 → 8 → 16 → 12 neurons (reconstruction)
  • Dropout: 0.2 dropout rate for regularization
  • Activation: ReLU for hidden layers, linear for output
• Ensemble Autoencoder: Multiple model voting system for improved accuracy
  • 3 different autoencoder configurations
  • Majority voting for anomaly classification
  • Confidence scoring for ensemble predictions
• Training Optimization:
  • Adam optimizer with learning rate 0.001
  • Mean Squared Error loss function
  • Early stopping with patience of 5 epochs
  • Batch size optimization (default 64)
• Anomaly Detection:
  • Statistical threshold calculation (mean + 1.5 × std)
  • Multi-level severity classification (Normal/Warning/Critical)
  • Feature contribution analysis for root cause identification
  • Major event pattern recognition

Application Layer (app.js)

• UI Management: Comprehensive interface for all user interactions
• Workflow Coordination: Orchestrates data fetching, training, and detection processes
• Real-time Visualization:
  • Training loss charts with validation curves
  • Reconstruction error timeline with color-coded severity
  • Anomaly severity distribution (doughnut charts)
  • Price and volume timeline with dual-axis display
• Data Visualization:
  • Raw data sample tables
  • Comprehensive statistics dashboard
  • Data quality metrics and warnings
  • Historical events analysis with October 10th special detection
• Model Persistence: Save/load functionality using IndexedDB and localStorage
• Memory Management: Proper tensor disposal and resource cleanup

Usage Instructions

1. Setup

1. Open index.html in a modern web browser
2. Ensure internet connection for Binance API access
3. No additional installation required (uses CDN libraries)

2. Data Collection

1. Select a trading pair from the dropdown (BTC/USDT, ETH/USDT, etc.)
2. Choose data interval (1h, 4h, 1d, 1w)
3. Configure date range:
   • Default: Use lookback days (default: 60 days)
   • Custom: Enable custom range and select start/end dates
4. Click "Fetch Market Data" to download historical data
5. Monitor batch processing progress for large date ranges
6. Review comprehensive data summary and quality metrics
7. Examine data sample table and statistics dashboard

3. Model Training

1. Adjust training parameters if needed:
   • Epochs: Number of training iterations (default: 30)
   • Batch Size: Training batch size (default: 64)
2. Click "Train Autoencoder" to begin training
3. Monitor real-time training progress and loss curves
4. Early stopping will automatically terminate training if no improvement
5. Training typically takes 1-3 minutes for standard model
6. Review training history and model performance metrics

4. Anomaly Detection

1. Click "Detect Anomalies" after training completes
2. Review the reconstruction error timeline chart with color-coded severity
3. Examine the anomaly severity distribution (doughnut chart)
4. Check the top anomalies table for specific timestamps and importance scores
5. Analyze feature contributions for the most significant anomalies
6. Review historical events analysis for major market events
7. Examine price and volume timeline for context

5. Model Management

• Save Model: Store trained weights and configuration locally using IndexedDB
• Load Model: Restore previously saved model and configuration
• Download Data: Export raw market data as JSON file
• Reset: Clear all data and start fresh with clean state

API Integration

Binance API Endpoints Used

• Klines: https://api.binance.com/api/v3/klines
  • Parameters: symbol, interval, startTime, endTime, limit
  • Returns: OHLCV data with comprehensive market information
  • Rate Limiting: 100ms delay between batch requests
  • Maximum Records: 1000 per request (Binance API limit)

Data Processing Pipeline

The application processes raw Binance data into twelve normalized features:

1. Price Change: Close-to-close percentage change
2. Volume: Normalized trading volume
3. Funding Rate Proxy: Simulated funding rate using price momentum and volume
4. Open Interest Proxy: Simulated open interest using volume and trade activity
5. Price Acceleration: Second derivative of price changes
6. Volume Spike: Detection of unusual volume increases (>30%)
7. Price Gap: Gap between opening price and previous close
8. Price Momentum: 3-period price momentum indicator
9. Volume Momentum: 3-period volume momentum indicator
10. RSI: 14-period Relative Strength Index
11. Bollinger Position: Position within Bollinger Bands (-1 to 1)
12. Market Regime: Market trend detection (uptrend/ranging/downtrend)

Anomaly Detection Logic

Threshold Calculation

• Training data reconstruction errors are analyzed using statistical methods
• Primary threshold: Mean + 1.5 × Standard Deviation
• Percentile-based threshold: 95th percentile of reconstruction errors
• Final threshold: Minimum of statistical and percentile methods for sensitivity
• This captures ~95% of normal market behavior while detecting major events

Severity Levels

• Green (Normal): Error < threshold
• Yellow (Warning): threshold ≤ error < 1.2 × threshold
• Red (Critical): error ≥ 1.2 × threshold

Feature Contribution Analysis

For each anomaly, the system calculates which features contributed most to the reconstruction error, helping identify the root cause of unusual market behavior. The analysis includes:

• Individual Feature Contributions: Absolute difference between original and reconstructed values
• Feature Importance Ranking: Sorted by contribution magnitude
• Pattern Recognition: Detection of specific anomaly patterns (crashes, spikes, etc.)
• Major Event Classification: Special flags for significant market events

Major Event Detection

The system includes specialized algorithms for detecting major market events:

• Crash Patterns: Price drop + volume spike combinations
• Flash Crashes: Extreme negative price changes with massive volume
• Market Manipulation: Unusual volume spikes with minimal price change
• Liquidation Cascades: Multiple consecutive anomalies with increasing severity

Interpretation Guide

Understanding Results

Reconstruction Error Timeline

• X-axis: Time points (chronological order)
• Y-axis: Reconstruction error magnitude
• Colors:
  • Green dots: Normal market conditions
  • Yellow dots: Elevated risk periods
  • Red dots: Critical anomalies
• Gray line: Anomaly threshold

Severity Distribution

• Shows percentage breakdown of normal vs. anomalous periods
• Helps assess overall market stability
• Higher anomaly rates may indicate volatile market conditions

Top Anomalies Table

• Lists timestamps of most significant anomalies
• Shows exact reconstruction error values
• Provides severity classification
• Enables correlation with external market events

Feature Contribution Analysis

• Identifies which market features drove the anomaly
• Price Change: Unusual price movements
• Volume: Abnormal trading volume
• Funding Rate Proxy: Momentum-based anomalies
• Open Interest Proxy: Activity-based anomalies

Common Anomaly Patterns

Liquidation Cascades

• High volume with negative price change
• Elevated funding rate proxy
• Multiple consecutive anomalies

Market Manipulation

• Unusual volume spikes with minimal price change
• Abnormal funding rate patterns
• Sudden reversals in price direction

Flash Crashes

• Extreme negative price changes
• Massive volume spikes
• Critical severity classification

Technical Specifications

Dependencies

• TensorFlow.js: 4.22.0 (neural network framework)
• Chart.js: 4.4.1 (data visualization)
• Native Fetch API: Binance API communication
• IndexedDB: Model persistence storage
• LocalStorage: Configuration and metadata storage

Browser Requirements

• Modern browser with ES6+ support
• WebGL support for TensorFlow.js GPU acceleration
• IndexedDB support for model persistence
• Minimum 4GB RAM recommended
• Chrome 80+, Firefox 75+, Safari 13+, Edge 80+

Performance Specifications

• Data Fetching:
  • Single request: ~2-5 seconds (up to 1000 records)
  • Batch processing: ~10-30 seconds (large date ranges)
  • Rate limiting: 100ms delay between batches
• Training Performance:
  • Standard model: 1-3 minutes (30 epochs)
  • Ensemble model: 3-9 minutes (3 models × 30 epochs)
  • Early stopping: Automatic termination after 5 epochs without improvement
• Anomaly Detection: <1 second for validation data
• Memory Usage:
  • During training: ~100-200MB
  • During inference: ~50-100MB
  • Tensor cleanup: Automatic disposal after operations

Data Processing Capabilities

• Maximum Data Points: 1000 per API request (Binance limit)
• Batch Processing: Automatic splitting for large date ranges
• Feature Engineering: 12 comprehensive market indicators
• Normalization: MinMax scaling across all features
• Data Quality: Automatic validation and error reporting

Model Architecture Details

• Input Features: 12-dimensional feature vectors
• Latent Space: 2-dimensional compressed representation
• Network Depth: 5 layers (encoder) + 5 layers (decoder)
• Total Parameters: ~1,000 trainable parameters
• Activation Functions: ReLU (hidden), Linear (output)
• Regularization: Dropout (0.2) and early stopping

User Interface Features

Modern Design Elements

• Glassmorphism Effects: Frosted glass appearance with backdrop blur
• Gradient Backgrounds: Professional blue-purple gradient theme
• Responsive Grid Layout: Adaptive design for all screen sizes
• Interactive Animations: Smooth hover effects and transitions
• Color-coded Elements: Consistent color palette for different data types

Advanced Controls

• Custom Date Range Selection: DateTime picker with validation
• Flexible Time Intervals: Support for multiple trading timeframes
• Training Parameter Adjustment: Real-time modification of epochs and batch size
• Data Range Options: Choice between custom dates or default lookback periods
• Model Management: Save/load functionality with persistent storage

Data Visualization Dashboard

• Real-time Progress Tracking: Live progress bars with percentage indicators
• Interactive Charts:
  • Training loss curves with validation metrics
  • Reconstruction error scatter plots with severity coloring
  • Anomaly distribution doughnut charts
  • Price/volume dual-axis timeline charts
• Data Quality Metrics: Comprehensive statistics with color-coded indicators
• Sample Data Tables: Preview of raw and processed market data
• Historical Events Analysis: Special detection for significant market events

User Experience Enhancements

• Status Messages: Real-time feedback on all operations
• Error Handling: User-friendly error messages with troubleshooting hints
• Loading States: Visual indicators during data processing and training
• Button State Management: Context-aware enabling/disabling of controls
• Memory Management: Automatic cleanup and resource optimization

Troubleshooting

Common Issues

"Failed to fetch market data"

• Check internet connection
• Verify Binance API is accessible
• Try different trading pair
• Check browser console for detailed errors

"Training failed"

• Ensure sufficient system memory
• Try reducing batch size or epochs
• Close other browser tabs
• Restart browser if memory issues persist

"No anomalies detected"

• Normal market conditions (good sign!)
• Try different time periods
• Adjust threshold sensitivity in code
• Check if training completed successfully

Charts not displaying

• Ensure Chart.js loaded correctly
• Check browser console for JavaScript errors
• Try refreshing the page
• Verify WebGL support

Performance Optimization

• Use smaller batch sizes on slower devices
• Reduce training epochs for faster results
• Close unnecessary browser tabs
• Use Chrome/Firefox for best performance

Future Enhancements

Planned Features

• Real-time Monitoring Mode: Live data streaming with WebSocket integration
• Multiple Timeframe Analysis: Simultaneous analysis across different intervals
• Custom Threshold Adjustment: User-configurable anomaly detection sensitivity
• Advanced Export Functionality: CSV, JSON, and PDF report generation
• Additional Trading Pairs: Support for more cryptocurrency pairs
• Advanced Feature Engineering: Technical indicators like MACD, Stochastic, etc.
• Model Comparison Tools: Side-by-side comparison of different model architectures

Technical Improvements

• WebSocket Integration: Real-time data streaming from Binance
• Model Ensemble Methods: Advanced voting and stacking techniques
• Automated Hyperparameter Tuning: Grid search and Bayesian optimization
• Cloud Deployment Options: Docker containers and cloud hosting
• Mobile-responsive Design: Enhanced mobile experience
• Progressive Web App: Offline functionality and app-like experience
• API Integration: RESTful API for external system integration

Advanced Analytics

• Pattern Recognition: Machine learning-based pattern identification
• Market Sentiment Analysis: Integration with social media and news data
• Risk Assessment: Portfolio-level risk analysis
• Backtesting Framework: Historical performance validation
• Alert System: Email and push notifications for anomalies
• Custom Dashboards: User-configurable visualization layouts

Contributing

This is an educational project demonstrating autoencoder-based anomaly detection in financial markets. Contributions and improvements are welcome.

Development Setup

1. Clone the repository
2. Open index.html in a local web server
3. Modify code as needed
4. Test thoroughly before deployment

License

This project is for educational purposes. Please ensure compliance with Binance API terms of service when using their data endpoints.

References

• Binance API Documentation
• TensorFlow.js Documentation
• Chart.js Documentation
• Autoencoder Theory: Hinton, G. E., & Salakhutdinov, R. R. (2006). Reducing the dimensionality of data with neural networks.