Technical Analysis Meets Machine Learning: Bitcoin Evidence

Battle Count: 62

Highly relevant to quantitative trading as it directly compares ML-based and TA-based trading strategies for Bitcoin. The paper provides practical evidence on strategy performance, includes transaction cost analysis, and addresses a real market event (Bitcoin ETF approval). However, it lacks advanced risk metrics (Sharpe, Sortino, max drawdown), does not implement a full trading system with position sizing, and the classification accuracy (~56%) is modest. The comparison framework and methodology are directly applicable to quantitative trading strategy development.

Moderate complexity. LSTM and LightGBM are well-established models with standard implementations available in popular libraries (TensorFlow/Keras, LightGBM package). Technical analysis strategies (EMA, MACD, ADX) are straightforward to implement. The main complexity lies in the hyperparameter grid search, proper train/test splitting, and the 95-day window management. No custom architectures or novel algorithms are proposed.

Data is publicly available on GitHub. Model architectures, hyperparameter grids, and training procedures are well-documented. Random seeds are set and results averaged over 10 initializations. However, exact software versions and computational environment details are not fully specified. The Binance data source is identified but raw data preprocessing steps could be more detailed.