Forecasting Cryptocurrency Prices Using Long Short-Term Memory

Abstract

Since the 1950s a discipline called ‘Artificial Intelligence’ has been gaining significant popularity. The curiosity about creating computers that can think and produce information like human beings has allowed scientists and computer engineers to contribute to this field. Many components such as robots, softwares and algorithms have been produced due to this purpose. Like various disciplines, Artificial Intelligence has been branched into several sub-disciplines. One of these branches is named ‘Machine Learning’. Machine Learning has different types of sub-branches such as Supervised Learning, Unsupervised Learning and Deep Learning. Deep Learning is the main Machine Learning technique used in this study. The ability to cope with complex situations allows Deep Learning models to be used in different application areas widespread. Predicting cryptocurrency prices can be counted as one of these applications. Because of investors’ desire to observe the cryptocurrency prices trend and reduce the investment risk using an effective method is becoming crucial. For this purpose, we created a Long Short-Term Memory which is a type of Deep Learning with the appropriate parameters via Python programming language. The dataset which is used to feed this model was obtained from the internet. After running the algorithm with this dataset, the validity of the model is calculated by a statistical tool called Mean Square Error. To visualize the effectiveness of the model’s output, a Python programming language library known as Matplotlib was chosen. Also, after the reviewing results of the model required interpretations and information about future studies will be explained by us in the Conclusion chapter.