Machine learning approaches for cryptocurrency price analysis

This study offers a thorough examination of various data formats and model setups for the short-term prediction of financial time series, presenting a deep learning framework based on Long Short-Term Memory (LSTM) enriched with technical indicators. The proposed model was tested on datasets prepared at both hourly and minute-level resolutions, where price-volume features and technical indicators were fed into the network through distinct input paths. Moreover, different model architectures namely LSTM, GRU, and BiLSTM were evaluated in combination with multiple loss functions including MSE, MAE, Huber, and a specially designed weighted loss function that applies stronger penalties to larger errors. The experimental findings indicate that the LSTM model, when supported by technical indicators, achieves high accuracy in predicting both the trend direction and magnitude of price movements. This research makes a valuable contribution to the field by adopting a holistic approach that integrates diverse data types and modeling strategies within the scope of deep learning-based financial forecasting.