Generative Pre-Trained Transformer Deep Learning And Sentimentanalysis-Based Bitcoin Cryptocurrency Price Prediction

Abstract

Predictability in the Bitcoin (BTC) prices can allay the fears of small investors. Existing techniques are yet to give importance to large language models and transformer architectures in developing the price prediction models. To fulfill the research gap, the current research paper has developed a model using Generative Pre-trained Transformer (GPT)architecture integrated with a linear artificial neural network (ANN) for making the finalprice prediction of BTCs in USD. The model is trained on BTC news articles dated June1, 2023, to June 1, 2024 from coindesk website collected through web scraping. Thenthe model was trained and after hyper-parameter tuning the results on the out-of-sample dataset were 5% mean absolute percentage error. Compared to state of the art models the performance of the proposed model on the other metrics like residual plots, meansquared error and prediction accuracy were higher by 2-5 %.