Abstract
Since the inception of Bitcoin (BTC) more than a decade ago, the popularity of cryptocurrency trading only exploded a few years ago when it reached the first all-time high in December 2017 at close to $20,000. As of 2019, there was an estimate of 51.2 to 52.4 million active cryptocurrency traders and an average of 2,900 to 3,000 cryptocurrencies available in the market (Bucquet et al., 2019). Even so, forecasting in the cryptocurrency exchange market remains a challenging task today even with the advancement of Artificial Intelligence (AI) and deep learning techniques. The reason is because many factors can affect the price and volatility in the market; on a broader level, it is being classified into technical and fundamental factors. This work explores the impact of different parameters such as time frame and sequence length in predicting cryptocurrency prices with deep learning models. The outcome is to build a model capable of accurately predicting immediate price movements based on past data. We do this by experimenting with various types of neural networks such as basic Long Short-Term Memory (LSTM), LSTM with Gated Recurrent Unit (GRU) and Bidirectional LSTM on Bitcoin’s candlestick chart coupled with different time intervals and sequence lengths to understand the relationship between these factors. The results of the experiments are analyzed to provide conclusive insights between the relationship of parameters followed by the implications of the paper’s findings on current state-of-the-art technologies used in the finance industry