Predicting and Enhancing the Fluctuations of Cryptocurrency Value by Using Machine Learning Algorithms

Neural network-based machine learning has applications in a wide range of industries, including translation, finance, distribution, the medical industry, and cognition. This paper demonstrates a recurrent neural network learning model based on LSTM that forecasts future Bitcoin prices by analysing past prices. By learning the previous prices for 30 days, this model predicts the next day’s price and shows the real and anticipated Bitcoin prices for 81 days. The training data set is divided at a rate of 1:9 from the regularised data set for modelling. The latter set is once more divided into verification and training data. This study’s machine learning requires the use of the Keras framework and Neural Network library. Using the training data, one must optimise the procedure in order to get the model’s weight. In this study, batch size for the fit function is 11 and epochs are 30. The loss declines more monotonously and eventually converges to a more regular value when learning is processed more frequently. In other words, it denotes that there is no overfitting. As a result of the experiment, machine learning not only suggests that weights converge towards a specific one after examining graphs of error rates and weight change rates, but also that as learning progresses, the neural network’s processing effectiveness improves.