Enhanced Bitcoin Price Prediction Using RNN-GRU Algorithm with Optimized Parameters: Overcoming LSTM Ambiguities for Improved Accuracy and Efficiency
M. Kathiravan, M. Meenakshi, M Kaviya, S. Sreesubha, V Sathyadurga, Vineet Gandhi
Abstract
Bitcoin's decentralised character and introduction in January 2009 can be mostly attributed to Satoshi Nakamoto, the unidentified designer of the digital money. Unlike regular money, Bitcoin has no value in and of itself and is not supported by any government or financial institution. The “mining” process consumes a lot of resources thus it cannot manage trades without a network of computers, often known as “nodes” or “miners.” The Bitcoin blockchain, a catalogue of all the past transactions, is maintained by these nodes. Like equities, bitcoin is getting more and more well-known even if it is somewhat erratic as an investment. Its value varies greatly; hence it is difficult to predict when prices will adjust. Since Bitcoin is so erratic, automated approaches are growing more crucial for future projections regarding it. Often utilised for prediction, long short-term memory (LSTM) networks have constraints. LSTM focusses on short-term input and has a complex and occasionally ambiguous design, thus some claim it does not operate well when prior data is not significant to current forecasts. This has caused some to wonder if one can forecast Bitcoin's price using this approach. Made as a substitute for standard RNNs, the Gated Recurrent Unit (GRU) helps to prevent the fading gradient issue. Bitcoin price projections made by the GRU design are far more accurate than those derived from more antiquated methods.