Distillation of Fuzzy Model by Feed Forward Neural Network for Cryptocurrency Trend Estimation

Author: Marek Ciklamini, and Josef Kokes.

ABSTRACT Deriving a fuzzy model (FM) where specific knowledge is stored into a common neural network can have various potentials. In this article, we extract common trading strategy into a fuzzy model which estimates trends (Fuzzy Trend Estimator, FZTE) of cryptocurrency, then a dataset is derived from the FZTE and cryptocurrency time series in order to be used for training simple architecture of feedforward neural network. We present a method to perform distillation of trading strategy compiled from a grey box model into a black box model. This allows us to investigate whether a combination of common mathematical modeling techniques is beneficial for creating new market signals.