محمدحسين صالحي نژاد

The growing popularity of digital currencies has made price prediction more promising for investors and researchers. With the advancement of artificial intelligence (AI) to such a level, cryptocurrency price prediction has become important. This research presents a price prediction system for any digital currency or any financial index, which is built using the most popular estimator algorithms based on deep learning. This research introduces an innovative method for estimating and predicting the value of digital currencies using the latest deep learning approaches based on recurrent neural networks. To do this thesis, we used Bitcoin, Ethereum and Cardano data in two-time frames, hourly and daily, which includes a set of 48 features at the beginning of training and learning. This data set is modeled by setting tomorrow's price prediction. The innovation of the thesis is to eva‎luate the price estimation in a hybrid model consisting of three PCA algorithms, automatic weighting based on correlation and without weighting. In price forecasting, we emphasize forecasting accuracy based on MAPE error reduction and apply advanced deep learning techniques. In order to establish a foundation for our proposed approach, we thoroughly review and compare the existing RNN methods for price estimation. Our method includes explaining the structure of the dataset, describing the preprocessing and implementation methods, and designing the best predictive model with the proposed algorithm. In the first step, in the combination of the modeling algorithm, we will examine the algorithms that have the best results for estimation and we will choose the best combination of the modeling algorithm. Then we put the selected algorithm as the base modeling algorithm and present the combined methods of data mining to combine the proposed final model. We focus on the final MAPE error. To present the results visually, we use charts and shapes implemented with PowerBI to give a proper understanding of the problem and show the ability of the algorithm to predict tomorrow's price. The implementation will be in Python language and using Tensorflow and pandas libraries.

پيش بيني قيمت , شبكه هاي عصبي بازگشتي , يادگيري ماشين , ارزهاي ديجيتال

Price Prediction , Recurrent Neural Networks , Machine Learning , Stock Price

Mohammad Hossein Salehinejad

Dr. Reza Entezari Maleki