Predicting Bitcoin Prices Using ANFIS and Haar Model

Abstract

This study aims to model and enhance the forecasting accuracy of cryptocurrency market data patterns using the daily bitcoin (BTC) close price data with 1535 observations from December 2017 to January 2022. The model employs a nonlinear spectral model of maximum overlapping discrete wavelet transform (MODWT) with Haar mathematical functions in conjunction with an adaptive network-based fuzzy inference system (ANFIS). We have selected the logarithm volume of bitcoin (LV) and logarithm trade count (LCT) as input values according to correlation and multiple regressions. The input and output variables have been collected from the cryptocurrency market. The performance of the proposed model (MODWT-Haar-ANFIS) is compared with traditional models that are the autoregressive integrated moving average (ARIMA) model and the ANFIS model. The obtained results show that the performance of MODWT-Haar-ANFIS is better than that of the traditional models. Therefore, the proposed forecasting model is a promising approach that capable of deploying in the cryptocurrency markets.