Crypto currency - Bitcoin prediction using machine learning techniques and news sources

In today's developing conditions, the development of technology and the internet, the emergence of various scientific developments have also affected money, which has an important place in people's lives. Money has taken its place as a different unit in the digital world for Bitcoin, a virtual currency. Bitcoin, which is one of the widely known virtual currencies today, is not affiliated to any institution or state, and its transfer is costless, easy and fast compared to real money, which increases the interest in Bitcoin all over the world day by day. This thesis examines the price prediction of Bitcoin, a cryptocurrency, using machine learning techniques. The study also considers news sources, social media data and market sentiment that affect Bitcoin price prediction and synchronizes these factors with machine learning models. Two-year datasets were created with tweets between 2021 and 2022 from https://www.kaggle.com and Bitcoin opening, closing, high, low and volume values between 2021 and 2022 from Yahoo Finance API. In machine learning with Python programming language, the sentiment states of tweets were evaluated using TextBlob, a natural language processing tool, and these sentiment states were combined with bitcoin price input data and the opening price of the next day was predicted with machine learning methods XGBoost, Random Forest, LSTM and Linear Regression. MAPE= 0.052624 error rate was found to be quite low. R² = 0.999999, a value close to 1, was found to have high accuracy in performance metrics. As a result of the study, it was observed that the Linear Regression algorithm gave more successful results. The application tested with real values that were not in the data set gave results close to the real values.