Predictive Analysis of Bitcoin Prices Using Bidirectional Long Short-Term Memory Networks

Abstract

Cryptocurrencies are a revolution in the domain of economics, currency, and trade, Bitcoin being the prime among them. With the ever increasing demand and limited number of bitcoins available, bitcoin as well as other cryptocurrencies has been a hot topic among economists, traders, inverters, and researchers. Although these coins based on blockchain are phenomenal, the volatile nature of the cryptocurrency markets has spurred significant interest in developing accurate prediction models to aid investors and market participants. This research paper examines the application of the Bidirectional Long Short-Term Memory model for predicting Bitcoin prices. This study uses historical Bitcoin price data and features such as opening prices, closing prices, and trading volumes to predict the target closing prices for Bitcoins. The study evaluates the performance of the Bi-LSTM model against comparative models, considering various relevant metrics. It was observed that Bi-LSTM performs significantly better among other regression models in capturing the inherent volatility and non-linearity of cryptocurrency markets. Our proposed model outperformed previous studies in various reported metrics. Additionally, we explored the impact of various hyperparameters and input features on the model’s performance to achieve an ideal state. Insights obtained by conducting this study would eventually contribute to an extensive understanding of the potential of deep learning techniques in forecasting cryptocurrency prices, offering valuable implications for investors and risk management strategies in the evolving landscape of digital assets.