A neural fuzzy network approach for Bitcoin price forecasting

Bitcoin is the most popular cryptocurrency built on blockchain technology infrastructure. It is a matter of great curiosity what the future price of Bitcoin, which has the largest volume in the crypto market under current conditions, will be. In this study, an attempt was made to predict the Bitcoin price direction. While making this prediction, a new data mining strategy called the 'collate method' was developed, which is the unique value of the study. In this context, the daily opening, closing, highest, lowest and volume values of the Bitcoin price were used as independent variables. The BTC/USD index data used in the study, which was analyzed with 3 separate data sets, was accessed from the Traindiview website. In the study, the 1st data set, which is the most comprehensive data set, covers the dates 17.07.2010-24.01.2024, which is the last date of analysis from the first day when Bitcoin's volume data was announced. The study was conducted with a sample of approximately 13.5 years. The 2nd data set covers the dates 03.01.2013-24.01.2024 and the 3rd period data set covers the dates 08.09.2019-18.04.2024. While trying to predict the Bitcoin price direction, analyzes were made using traditional and blending methods. Afterwards, accuracy rate analysis was performed. Finally, Root Mean Square Error (RMSE) values were analyzed for performance measurements of our findings. The findings indicate that the developed blending method provides a higher prediction of Bitcoin's spot prices than the traditional method. According to the accuracy classification rate findings; It determines whether the model classifies the value it found as a decline for the day that occurred as a decline day or as a decline for the day that occurred for an increase. In RMSE findings, it was determined that as the percentage of correct prediction success increased, error values decreased.