System Architecture

Overview

CryptoView Pro is a sophisticated cryptocurrency forecasting system built with a modular architecture that combines real-time data collection, technical analysis, machine learning predictions, and user notifications. The system is designed to provide accurate short-term and long-term price predictions using hybrid ML models.

System Components

The application is organized into several key modules:

Data Layer

Handles data collection from cryptocurrency exchanges using CCXT

ML Layer

Three specialized models: XGBoost, Prophet, and Hybrid

Analysis Layer

Technical indicators (RSI, MACD, Bollinger Bands, EMAs)

Presentation Layer

Streamlit-based interactive dashboard

Architecture Diagram

Core Components

1. Data Collection Layer

Location: data/collectors.py The CryptoDataCollector class manages all data retrieval operations:

class CryptoDataCollector:
    def fetch_ohlcv(symbol, timeframe, limit):
        """Fetches OHLCV data with caching"""
        # Returns: timestamp, open, high, low, close, volume
        # Adds: returns, log_returns
        
    def get_current_price(symbol):
        """Real-time price retrieval"""
        
    def get_24h_stats(symbol):
        """24-hour statistics and metrics"""

Features:

Supports multiple exchanges via CCXT (default: Kraken)
Built-in caching with configurable TTL (60 seconds)
Automatic rate limiting
Error handling and retry logic
Calculates returns and log returns automatically

Data Flow:

User selects cryptocurrency and timeframe in sidebar
CryptoDataCollector.fetch_ohlcv() queries exchange API
Raw OHLCV data is cached in Streamlit session
Data is enriched with calculated returns
DataFrame is passed to technical indicators layer

2. Technical Analysis Layer

Location: utils/indicators.py The TechnicalIndicators class provides comprehensive technical analysis:

class TechnicalIndicators:
    @staticmethod
    def calculate_rsi(data, periods=14):
        """Relative Strength Index"""
        # Measures momentum (0-100 scale)
        
    @staticmethod
    def calculate_macd(data, fast=12, slow=26, signal=9):
        """Moving Average Convergence Divergence"""
        # Returns: macd, signal_line, histogram
        
    @staticmethod
    def calculate_bollinger_bands(data, period=20, std_dev=2):
        """Volatility bands"""
        # Returns: upper, middle, lower
        
    @staticmethod
    def calculate_ema(data, period):
        """Exponential Moving Average"""
        # Computed for periods: 9, 21, 50, 200

Signal Generation: The system generates trading signals from technical indicators:

RSI Signals: Overbought (>70), Oversold (<30)
MACD Signals: Bullish (MACD > Signal), Bearish (MACD < Signal)
Overall Signal: Aggregates multiple indicators for buy/sell/neutral

3. Machine Learning Layer

Location: models/ Three specialized predictive models work together:

XGBoost
Prophet
Hybrid

Best for: 1-72 hours (short-term trading)

Gradient boosted decision trees
60+ engineered features
Fast training and inference
High accuracy for intraday predictions

See Models for details.

4. Visualization Layer

Location: app.py (chart functions) Three main chart types are generated:

Main Price Chart

Candlestick chart with volume subplot

OHLCV candlesticks
EMA overlays (50, 200)
Bollinger Bands
Volume bars with direction coloring
Time range selectors (24H, 2D, 7D, All)

Prediction Chart

Historical data with future forecasts

Last 336 hours of historical context
Predicted price line with confidence intervals
“Now” marker showing current time/price
Model comparison mode (optional)

Technical Indicators Chart

Dedicated charts for RSI and MACD

RSI with overbought/oversold lines
MACD with signal line and histogram
Unified hover mode

5. Alert System

Location: utils/alerts.py, utils/telegram_notifier.py Multi-channel alerting system:

# Alert types supported
alert_types = [
    "Precio",      # Price threshold
    "RSI",         # RSI overbought/oversold
    "Cambio %"     # Percentage change
]

# Conditions
conditions = ["Mayor que", "Menor que"]

# Delivery channels
- In-app notifications
- Telegram messages (if configured)

Alert Flow:

User creates alert with condition/threshold
Alert stored in st.session_state.alerts
Background check evaluates conditions
If triggered, notification sent via Telegram
Alert displayed in dashboard

Data Flow

Complete Request Cycle

Session State Management

The application maintains state across interactions:

st.session_state.collector         # CryptoDataCollector instance
st.session_state.data             # Current OHLCV DataFrame
st.session_state.predictor_xgb    # XGBoost model instance
st.session_state.predictor_prophet # Prophet model instance
st.session_state.predictor_hybrid  # Hybrid model instance
st.session_state.model_trained    # Boolean training flag
st.session_state.predictions      # Latest predictions
st.session_state.backtest_results # Validation metrics
st.session_state.selected_model   # Current model choice
st.session_state.alerts           # Active alerts list
st.session_state.telegram_notifier # Telegram client

Configuration System

Location: config/settings.py Centralized configuration using environment variables:

API Configuration

BINANCE_API_KEY = os.getenv('BINANCE_API_KEY', '')
BINANCE_API_SECRET = os.getenv('BINANCE_API_SECRET', '')

Cryptocurrency Settings

AVAILABLE_CRYPTOS = [
    'BTC/USDT', 'ETH/USDT', 'BNB/USDT', 'SOL/USDT',
    'ADA/USDT', 'XRP/USDT', 'DOGE/USDT', 'DOT/USDT',
    'MATIC/USDT', 'AVAX/USDT', 'LINK/USDT', 'UNI/USDT'
]

TIMEFRAMES = {
    '1m': '1 minuto', '5m': '5 minutos', '15m': '15 minutos',
    '1h': '1 hora', '4h': '4 horas', '1d': '1 día'
}

Model Parameters

MODELS_CONFIG = {
    'xgboost': {
        'n_estimators': 200,
        'learning_rate': 0.07,
        'max_depth': 6,
        'subsample': 0.8,
        'colsample_bytree': 0.8
    },
    'prophet': {
        'changepoint_prior_scale': 0.5,
        'seasonality_prior_scale': 10,
        'daily_seasonality': True,
        'weekly_seasonality': True
    }
}

Technical Indicators

INDICATORS_CONFIG = {
    'rsi_period': 14,
    'macd_fast': 12,
    'macd_slow': 26,
    'macd_signal': 9,
    'bb_period': 20,
    'bb_std': 2,
    'ema_periods': [9, 21, 50, 200]
}

Performance Optimizations

Caching

Streamlit’s @st.cache_data decorator caches:

Exchange API calls (60s TTL)
Technical indicator calculations
Model predictions

Lazy Loading

Models are loaded only when needed:

XGBoost imported conditionally
Prophet imported conditionally
Fallback to available models

Vectorization

NumPy/Pandas operations:

Vectorized indicator calculations
Batch predictions
Efficient data transformations

Session Persistence

Models persist in session:

Train once, predict many times
Reuse scalers and feature sets
Avoid redundant computations

Error Handling

Robust error handling at every layer:

API Errors

Exchange connection failures handled gracefully with user-friendly messages

Data Validation

Minimum data points enforced (500+), empty DataFrames detected

Model Errors

Training failures caught with informative error messages

Feature Engineering

NaN values from indicators are dropped automatically

Security Considerations

API keys and secrets are loaded from environment variables (.env file) and never committed to version control.

The system uses read-only API access for data collection. No trading operations are performed.

Scalability

The architecture supports scaling in several dimensions:

Horizontal: Multiple instances can run independently
Data Sources: Easy to add new exchanges via CCXT
Models: New ML models can be integrated with minimal changes
Cryptos: Simply add symbols to AVAILABLE_CRYPTOS list
Indicators: Extend TechnicalIndicators class with new calculations

Get Started

Core Concepts

ML Models

Features

Configuration

Overview

System Components

Data Layer

ML Layer

Analysis Layer

Presentation Layer

Architecture Diagram

Core Components

1. Data Collection Layer

2. Technical Analysis Layer

3. Machine Learning Layer

4. Visualization Layer

5. Alert System

Data Flow

Complete Request Cycle

Session State Management

Configuration System

Performance Optimizations

Caching

Lazy Loading

Vectorization

Session Persistence

Error Handling

Security Considerations

Scalability

Next Steps

ML Models

Technical Indicators

Build docs developers (and LLMs) love

Get Started

Core Concepts

ML Models

Features

Configuration

​Overview

​System Components

Data Layer

ML Layer

Analysis Layer

Presentation Layer

​Architecture Diagram

​Core Components

​1. Data Collection Layer

​2. Technical Analysis Layer

​3. Machine Learning Layer

​4. Visualization Layer

​5. Alert System

​Data Flow

​Complete Request Cycle

​Session State Management

​Configuration System

​Performance Optimizations

Caching

Lazy Loading

Vectorization

Session Persistence

​Error Handling

​Security Considerations

​Scalability

​Next Steps

ML Models

Technical Indicators

Build docs developers (and LLMs) love

Overview

System Components

Architecture Diagram

Core Components

1. Data Collection Layer

2. Technical Analysis Layer

3. Machine Learning Layer

4. Visualization Layer

5. Alert System

Data Flow

Complete Request Cycle

Session State Management

Configuration System

Performance Optimizations

Error Handling

Security Considerations

Scalability

Next Steps