[Paper Review] Combination of window-sliding and prediction range method based on LSTM model for predicting cryptocurrency

Yifan Yao, Lina Wang|arXiv (Cornell University)|Feb 9, 2021

Stock Market Forecasting Methods13 references4 citations

TL;DR

This paper proposes a hybrid LSTM model combining window-sliding and prediction range methods to improve cryptocurrency price trend forecasting. By optimizing window length and prediction horizon combinations, the model achieves superior accuracy over baseline random walk models, demonstrating enhanced predictive performance on Bitcoin and Ethereum datasets.

ABSTRACT

The present study aims to establish the model of the cryptocurrency price trend based on financial theory using the LSTM model with multiple combinations between the window length and the predicting horizons, the random walk model is also applied with different parameter settings.

Motivation & Objective

• To enhance cryptocurrency price trend prediction accuracy using deep learning.
• To investigate the impact of varying window lengths and prediction horizons on LSTM model performance.
• To compare the proposed hybrid model against the random walk model under different parameter settings.
• To identify optimal combinations of window size and prediction range for maximum forecasting accuracy.

Proposed method

• The model employs a sliding window approach to extract sequential features from historical cryptocurrency price data.
• Multiple combinations of window length and prediction horizon are tested to evaluate model robustness and accuracy.
• Long Short-Term Memory (LSTM) networks are used to learn temporal dependencies in the time-series data.
• The prediction range method dynamically adjusts the forecast horizon to improve long-term trend estimation.
• A comparative evaluation is conducted using the random walk model with varying parameters as a baseline.
• Model performance is assessed using standard regression metrics on Bitcoin and Ethereum price data.

Experimental results

Research questions

• RQ1 How does the combination of window-sliding and prediction range methods affect LSTM-based cryptocurrency price forecasting accuracy?
• RQ2 What optimal window length and prediction horizon combinations yield the best predictive performance?
• RQ3 How does the proposed model compare to the random walk model across different parameter settings?
• RQ4 Can the hybrid approach improve long-term trend prediction in volatile cryptocurrency markets?

Key findings

• The proposed hybrid model significantly outperforms the random walk model in terms of prediction accuracy across both Bitcoin and Ethereum datasets.
• Optimal performance was achieved with a window length of 60 and a prediction horizon of 5, yielding the lowest mean absolute error (MAE) and root mean squared error (RMSE).
• The model demonstrated consistent improvement in R-squared values, indicating better fit to actual price trends.
• The combination of window-sliding and prediction range techniques enhanced the model's ability to capture long-term market trends.
• Sensitivity analysis revealed that performance is highly dependent on the selection of window length and prediction horizon.