Bitcoin Price Prediction using Machine Learning and Combinatorial Fusion Analysis

Battle Count: 315

Highly relevant to quantitative trading as it directly addresses next-day Bitcoin price prediction, which is fundamental for trading decisions (buy/hold/sell). The CFA ensemble approach demonstrates significant improvement over individual models (MAPE of 0.19% vs 1.20-4.91% for base models). The methodology could be adapted for other financial instruments. However, the paper focuses on point prediction rather than trading strategy backtesting, position sizing, or risk-adjusted returns.

Moderate complexity. Requires implementing five diverse ML models (SVM, RF, XGBoost, CNN, LSTM), generating prediction distributions, computing RSC functions and cognitive diversity, and performing 104 combination evaluations. The CFA framework itself is well-defined mathematically. Hyperparameter tuning via 10-fold cross-validation and random search adds computational overhead. The sliding window approach for time series adds implementation complexity.

The paper provides detailed methodology descriptions including data sources, feature engineering, model architectures, and CFA combination strategies. However, no GitHub repository or code link is mentioned. Hyperparameter tuning details (10-fold cross-validation, random search) are described but specific parameter values are not fully listed. Data sources are publicly available (Coinmetrics, Yahoo Finance, Blockchain.com).