Tuning the Accuracy in the Price Prediction of Cryptocurrency Using Long Short Term Memory in Comparison with XG Boost

Cryptocurrency is one of the favorite investment options, and it remains a challenge to predict the price of the cryptocurrency correctly due to high market volatility. There are several ways to predict the price of cryptocurrency; however, most of the traditional techniques have failure points in capturing the peculiar patterns available within the data of cryptocurrency. The aim of this study is to enhance cryptocurrency price prediction by comparing two techniques, Long Short-Term Memory and XGBoost. In this paper, we compared the performance of these two methods against each other. We have taken 1,684 samples for the study; we have used 1,200 to train the models and 594 for testing purposes. Our objective was to predict cryptocurrency prices at each sample size 10, using a threshold of 0.05 % on a 95 % confidence interval. The results have shown that the LSTM model has a very high accuracy of 94.405 %, while the XGBoost model stands at 85.261 %. We have also obtained a p-value of 0.002 ((<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$p; 0.05$</tex>)), indicating that the results are statistically significant to detect accurate price predictions. This study therefore postulates that in cryptocurrency price prediction, the use of Long Short-Term Memory is preferable to the use of XGBoost, considering the high accuracy.