An Empirical Analysis on Price Forecasting of Bitcoin Using Deep Learning Models

Abstract

Cryptocurrencies are among the most actively traded financial assets in the world, making price prediction a key concern for investors today. Currently, one kind of stock market investing is Bitcoin, a sort of cryptocurrency. Numerous things impact the stock market. Among the many cryptocurrencies that have surged in value recently is Bitcoin, which sometimes drops sharply without anybody realizing its impact on the stock market. An automated technique to forecast Bitcoin on the stock market is required due to its swings. The dataset, which was preprocessed, analyzed, and visualized using cryptocurrencies, was utilized for this study. The study suggests a CNN-GRU model that uses Gated Recurrent Units for sequence prediction and Convolutional Neural Networks for feature extraction. The current literature on price prediction is fragmented, inaccurate, and focused mostly on short-term forecasts.  Training and testing the CNN-GRU model on the existing dataset to generate a long-term prediction is the innovative aspect of this study. Some measures of performance include decision accuracy, R2 Score, MSE, RMSE, and MAE. As a comparison, the best training performance for period 2 is 0.9988 with an RMSE of 562.08, MAPE of 0.0451, and R2 score of 0.9907. On the other hand, the best testing performance for period 2 is 0.9907 with an RMSE of 846.18, MAPE of 0.0134. Although it requires more time to build, the CNN-GRU model is determined to be the superior mechanism for time-series Bitcoin price prediction.