A Deep Learning Approach to Forecast Cryptocurrency Prices

EasyChair Preprint 15701

Abstract

This work aims to propose deep learning technique that combines convolutional neural network with single multiplicative neuron model to optimize delay value and improving forecasting efficiency in predicting cryptocurrency
prices. This model is proposed with the intent of tackling high non-linearity present in the cryptocurrency prices.
A uni-variate time series of daily price of two cryptocurrencies Bitcoin and Ethereum is considered to validate the
proposed model. Multiple experiments have been performed to validate the proposed deep learning model and RMSE
value is used as the error criteria. The least RMSE value is used in evaluating optimal delay value. The proposed
model is 23%-33% is more accurate in forecasting compared to the single multiplicative neuron model. The results
obtained can give valuable insights for decision making. This work will enable future research studies in time series
prediction, as well as facilitate easy adaptation to various time series and with different scenarios.

Keyphrases: Convolutional Neural Network, Cryptocurrency, deep learning, forecasting cryptocurrency prices, time series

@booklet{EasyChair:15701,
  author    = {Shobhit Nigam},
  title     = {A Deep Learning Approach to Forecast Cryptocurrency Prices},
  howpublished = {EasyChair Preprint 15701},
  year      = {EasyChair, 2025}}