New Fuzzy Approaches to Cryptocurrencies Investment Recommendation Systems

Abstract

This work proposes the use of Computational Intelligence algorithms to predict cryptocurrencies values based on historical values. After predicting the value of the currencies for up to three days following the current one using an evolving algorithm, two approaches were presented to suggest the investment: The first one uses only the result of the forecast to provide a suggestion of investment; in contrast, the second approach, in addition to using the prediction data returned by the evolving system, also applies a Mamdani system, based on expert knowledge, to suggest to the users what to do with their invested value. After performing and processing the historical data of three cryptocurrencies, the suggestions offered by both approaches were compared to the actual quote. The comparison presented results with a total assertiveness rate of over 90% for the three cryptocurrencies evaluated, according to established criteria, for both the evolving approach and the hybrid approach. Computational experiments suggest that the two proposed approaches are promising and competitive with alternatives reported in the literature.