A Comparison of Long Short-Term Memory and Recurrent Neural Network for Novel Cryptocurrency Price Prediction with a View to Improving Accuracy

Main Article Content

K. Pavanesh, John Justin Thangaraj


Aim: The aim of this paper is to achieve better accuracy among the pair of bitcoin prices in USD that can be compared and predicted. The price of crypto data is obtained from the source of Bitcoin Price Index. The task is achieved with fluctuating levels of progress through the implementation of a Recurrent Neural Network organization (RNN) also, a Long Short-Term Memory (LSTM) Network. Materials and Methods: LSTM with RNN is a widely used neural network algorithm for predicting bitcoin price as it remembers some important data which is received by the input and helps them to predict price of the next output accurately. It is mostly used in sequential data. The LSTM algorithm is applied with sample size = 750 and RNN algorithm is computed with sample size = 750, where evaluated many times with the computation for number of iterations N=10 to predict the accuracy percentage. Results: LSTM algorithm has better accuracy (85.60%) when compared to RNN accuracy (76.70%). The statistical significant value (two-tailed) is 0.001 (p<0.01). Each group consists of a sample size of 10 and the study parameters include alpha value 0.05, beta value 0.2, and the power value 0.8. The sample size was calculated using G Power value as 80%. Conclusion: The study proves that Long Short-Term Memory exhibits better accuracy than Recurrent Neural Network in predicting the Cryptocurrency price prediction.

Article Details