Unsupervised Clustering and Ensemble Decision Strategies in Cryptocurrency Trading

A SOM-Based Hybrid Model for Signal Generation

This project applies Self-Organizing Maps (SOMs) to generate and evaluate trading signals based on technical indicators, sentiment data, or a hybrid of both. It supports training, evaluation, ensemble strategies, and backtesting.

Figure 1. Methodological framework

Project Structure

.  
├── data/                  # Input data: price, sentiment  
├── ensemble/              # Strategies combining multiple SOMs  
├── evaluation/            # Evaluation metrics and backtesting tools  
├── pipeline/              # Data preprocessing and feature engineering  
├── train/                 # SOM training scripts and utilities  
├── config.py              # Global configuration  
├── main.ipynb             # Main execution notebook (demo/workflow)  
├── requirements.txt       # Required Python packages  

Install dependencies:

   pip install -r requirements.txt

Use main.ipynb to run the full pipeline, from feature selection and SOM training to signal generation and strategy evaluation.