Abstract: The rise and subsequent crash of the blockchain era have transformed cryptocurrencies into investment assets, necessitating accurate forecasts to guide investment decisions due to their highly unpredictable nature. While existing studies have utilized machine learning to predict Bitcoin prices with improved accuracy, few have explored the applicability of different modelling techniques to diverse data formats and dimensional attributes. This research focuses on categorizing Bitcoin prices into daily and high-frequency intervals, with the goal of anticipating cryptocurrency prices at various frequencies using machine learning techniques. For daily price prediction, a comprehensive set of high-dimensional aspects, including property and network characteristics, trading and market indicators, attention metrics, and gold spot prices, are leveraged. On the other hand, 5-minute interval price prediction relies on fundamental trading data obtained from cryptocurrency exchanges. Given the influence of major organizations on price control and the volatile nature of the cryptocurrency market, precise forecasting methods that consider factors such as market capitalization, maximum supply, volume, and circulating supply are essential. Deep learning techniques, such as recurrent neural networks (RNNs), long short-term memory (LSTM), and gated recurrent units (GRU), are employed as effective models for training the data. The proposed approach is evaluated using benchmark datasets and implemented in Python. The results demonstrate the efficacy of the suggested methodology in achieving accurate predictions. Consequently, neural networks, as intelligent data mining technologies, have gained widespread adoption in various sectors over the past decade, offering valuable insights into cryptocurrency price forecasting.

Keywords: LSTM, RNN, Cryptocurrency, blockchain.

PDF | DOI: 10.17148/IJARCCE.2023.12682

Open chat
Chat with IJARCCE