Bitcoin Cryptocurrency Price Prediction Using IFA-BiLSTM

Abstract

Bitcoin stands out as the most valuable asset in the cryptocurrency market, characterized by its highly fluctuating and unpredictable price. Investments relying on price fluctuations entail high levels of risk, thus requiring accurate methods to predict Bitcoin price estimates. This research aims to devise a solution for predicting the price value of Bitcoin with fluctuating price characteristics using a combination of the Improved Firefly Algorithm (IFA) and Bidirectional Long Short Term Memory (BiLSTM) methods. The predicted data is preprocessed through min-max normalization before being split into training and testing sets, maintaining an 80:20 ratio. Then, the IFA method will search for the best hyperparameter values for BiLSTM. By using the right hyperparameter values that are appropriate for the data, the model can generate good prediction accuracy. The IFA-BiLSTM model outperforms previous models with an RMSE of 2051.55, MAPE of 3.48%, and accuracy of 96.52%.