Predicting Bitcoin Price Direction Using Machine Learning Models

Abstract

In the financial sector, as past economic and social events have shaken trust, this trust is being regained through the internet and computer technologies. Emerging in the 19th century, financial technology has led to a new economic understanding with digital money and especially bitcoin. The decentralized structure of bitcoin and the encryption systems used for security play an important role in preventing fraud and have become the center of attention of investors. As its value has increased, studies on price predictions have naturally increased. This study aims to predict the impact of data obtained from digital economy news sites on bitcoin price using natural language processing and machine learning techniques. In line with this goal, text vectorization was performed with the TF-IDF statistical method. Synthetic Minority Oversampling Technique (SMOTE) was applied to eliminate the imbalance in the vectorized data set. Classification models such as Logistic Regression, Decision Trees, Random Forest, Support Vector Machines, K-Nearest Neighbor, Extra Trees, Bernoulli Naive Bayes and Multilayer Perceptron were applied to the obtained output.According to the results of the performance of different machine learning models in predicting the direction of bitcoin price fluctuation, the Extra Trees Classifier model showed the highest performance with an Accuracy of 86.71%, recall of 86.71%, precision of 86.99% and F1 score of 86.59%.