A carregar...
Publicação
Unsupervised Clustering and Ensemble Decision Strategies in Cryptocurrency Trading: A SOM-Based Hybrid Model for Signal Generation
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.06 MB | Adobe PDF |
Orientador(es)
Resumo(s)
This study presents a hybrid financial modeling framework that combines technical indicators
and sentiment analysis to generate trading signals for the BTC-USD market using SelfOrganizing Maps (SOMs). The proposed pipeline integrates two data sources: numerical price
data and financial news, from which sentiment scores are extracted using FinBERT. After
feature engineering and normalization, three SOMs are trained independently, one using only
technical features, one using only sentiment features, and one combining both. A
comprehensive grid search is performed to optimize the feature selection and SOM
hyperparameters. The resulting cluster assignments are translated into trading signals (buy,
hold, sell) and evaluated using four ensemble strategies: majority, unanimous, weighted, and
aggressive voting, and, among these, the weighted ensemble is further optimized by testing
various weight combinations for each signal source. The framework is tested over a defined
out-of-sample period, and its performance is assessed using metrics such as cumulative
return, sharpe ratio, and maximum drawdown. The results demonstrate that combining SOMbased clustering with ensemble decision strategies can yield interpretable and competitive
trading signals in volatile markets like cryptocurrency. This work highlights the relevance of
unsupervised learning techniques and multi-source data integration in financial forecasting
and trading automation.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Bitcoin Ensemble Learning Sentiment Analysis Self-Organizing Maps Technical Indicators SDG 8 - Decent work and economic growth