Comparative Study on Neural Networks Models for Bitcoin Price Prediction Using Technical Indicators and Bayesian Optimization

Abstract

Bitcoin, the most widely used cryptocurrency, has garnered significant interest due to its volatile nature and the challenges associated with its price prediction. This paper presents a comparative study of various neural network algorithms for Bitcoin price prediction, utilizing daily trading data and multiple technical indicators. The models evaluated include Artificial Neural Networks (ANN), Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), Backpropagation Neural Networks (BPNN), and Long Short-Term Memory (LSTM). A baseline comparison is conducted using Linear Regression (LR). Bayesian Optimization is employed to enhance the performance of these models by fine-tuning their hyperparameters. The study aims to determine the most effective model for accurate Bitcoin price prediction, which is crucial for both investors and market analysts. The results indicate that selective use of technical indicators and Bayesian Optimization significantly improves the MSE and RMSE scores of most models in this paper, compared to using only a base model without technical indicators.