CryptoPred

Real-time cryptocurrency price prediction platform powered by machine learning

Features • How It Works • Tech Stack • User Guide • Technical Docs

---

Overview

CryptoPred is an end-to-end machine learning platform for cryptocurrency price prediction. It processes real-time market data, computes technical indicators, analyzes social sentiment, and generates price forecasts using state-of-the-art ML models.

Built for traders, researchers, and fintech teams who need reliable, explainable predictions at scale.

---

Key Features

Real-Time Data Processing

Stream trades from major exchanges with sub-second latency. Automatic aggregation into OHLCV candles at multiple timeframes.

Advanced Technical Analysis

40+ technical indicators computed in real-time: moving averages, RSI, MACD, Bollinger Bands, volume profiles, and more.

Social Sentiment Integration

LunarCrush integration for social media sentiment scores, galaxy scores, and market correlation metrics.

Machine Learning Models

Production-ready LightGBM and ensemble models with automatic hyperparameter optimization and model versioning.

Drift Detection

Automatic detection of data distribution shifts and model degradation with alerting.

Full Observability

Grafana dashboards for monitoring data pipelines, model performance, and prediction quality.

---

How It Works

┌─────────────┐    ┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│  Exchanges  │───▶│   Kafka     │───▶│ RisingWave  │───▶│   MLflow    │
│  (Binance)  │    │  (Streams)  │    │ (Features)  │    │  (Models)   │
└─────────────┘    └─────────────┘    └─────────────┘    └─────────────┘
                                                               │
┌─────────────┐    ┌─────────────┐    ┌─────────────┐          │
│ LunarCrush  │───▶│  Sentiment  │───▶│  Combined   │◀─────────┘
│    (API)    │    │  Features   │    │ Predictions │
└─────────────┘    └─────────────┘    └─────────────┘

1. Data Ingestion — Real-time trade streams from Binance WebSocket API
2. Feature Engineering — Technical indicators and sentiment features computed via streaming SQL
3. Model Training — Periodic retraining with Optuna hyperparameter search
4. Prediction — Continuous inference with confidence intervals
5. Monitoring — Drift detection and performance tracking

---

Tech Stack

Component	Technology
Language	Python 3.13
ML Framework	LightGBM, scikit-learn, Optuna
Streaming	Apache Kafka (Strimzi)
Feature Store	RisingWave (streaming SQL)
MLOps	MLflow
Orchestration	Kubernetes
Monitoring	Grafana, Prometheus
Data Source	Binance API, LunarCrush API

---

Supported Trading Pairs

Exchange	Pairs	Timeframes
Binance	BTC/USDT, ETH/USDT, SOL/USDT	1m, 5m, 15m, 1h

Additional pairs can be configured via environment variables.

---

Model Performance

Metric	Value
RMSE (5-min horizon)	< 0.5%
Direction Accuracy	> 55%
Retraining Frequency	Daily
Inference Latency	< 100ms

Metrics based on backtesting. Past performance does not guarantee future results.

---

Architecture Highlights

• Streaming-first — All data processing uses Apache Kafka and RisingWave streaming SQL
• Cloud-native — Kubernetes-native with Helm charts for production deployment
• MLOps-ready — MLflow integration for experiment tracking and model registry
• Observable — Prometheus metrics, Grafana dashboards, structured logging

---

Documentation

Document	Description
User Guide	Setup, configuration, and daily operations
Technical Documentation	Architecture, API reference, development guide

---

Roadmap

See ROADMAP.md for the detailed development plan including:

• On-chain analytics & derivatives data
• Deep learning models (LSTM, Transformer, TFT)
• Uncertainty quantification & conformal prediction
• Backtesting framework with strategy simulation
• Trading signal generation & risk management
• REST/WebSocket API for predictions
• Web dashboard & alerting integrations

---

License

This project is proprietary software. All rights reserved.

---

_{Built with Python, Kafka, RisingWave, LightGBM, and MLflow}