A Novel Method of Bitcoin Price Prediction

Abstract

The popularity and value of Bitcoin have attracted numerous traders and researchers, leading to countless studies on prediction of Bitcoin price. To improve accuracy, relevant features with strong correlation to Bitcoin prices are typically chosen from the dataset, and a random data subset is used for model training and model testing. Unfortunately, the random selection of data may lead to unfitting outcomes, thereby reducing the accuracy of the price predictions. To look into it, a methodology is proposed that involves identifying suitable data chunks for model training. Using this methodology, Simple Long Short Term Memory model is trained to predict future five days Bitcoin prices. The outcome showed that when LSTM model is trained with an appropriate data block, sustainable results were achieved. Overall, the study highlights importance of carefully selecting machine learning algorithms, training data, exceptionally for Bitcoin price prediction for future five days.