Comparative Performance Analysis of Deep Learning Models for Cryptocurrency Price Forecasting

Abstract

Purpose: This research aims to find an accurate cryptocurrency price prediction model to mitigate financial risks caused by high price volatility. This research compares the predictive capabilities of five Deep Learning model, namely LSTM, GRU, BiLSTM, Transformer, and Performer, for  predicting cryptocurrency prices with the highest accuracy in the digital financial market.

Methods: The methods applied in this research are dataset, preprocessing data, model training, and model evaluation. The dataset used in this study, namely the price per minute data for BTC, ETH, BNB, and XRP, was obtained from Kaggle. Data processing includes normalization using MinMaxScaler and sequence generation through the Sliding Window technique. To validate each deep learning model, and four metrics consisting of MAE, MSE, RMSE, and MAPE are used for evaluation.

Result: The Transformer model created the best results for the lowest MAPE value across all datasets, the smallest being BTC and ETH at 0.20%, BNB at 0.29%, and XRP at 0.36% demonstrating high accuracy and generalization. The BiLSTM was ranking second since it captured effectively the bidirectional temporal dependencies; the GRU was moderate but stable in its performance. The data showed that the accuracy of LSTM and Performer varied.

Novelty: This study offers a comprehensive comparison of various Deep Learning models in detail, enabling it to find the best model for predicting cryptocurrency prices with high accuracy. This study provides valuable insights for the development of advanced deep learning-based price forecasting systems in the field of digital financial analysis.