Research of the application of automated machine learning for the comparative analysis of cryptocurrency rate forecasting methods

Abstract

The forecasting of asset values has always attracted significant interest, prompting researchers to continuously refine methods and tools for addressing this task. The advancement of artificial intelligence has greatly enhanced research potential in this field, leading to the emergence of new algorithms and methods that leverage the advantages of computational speed and the accuracy of neural networks. This has fueled active research in machine learning for forecasting, resulting in the development of new variations and modifications of algorithms. However, the abundance of research necessitates users to independently analyze and experimentally verify their effectiveness. This article explores the possibility to apply automated machine learning (AutoML) to analyze and compare methods for cryptocurrency price predictions. At the time of the research, AutoML is mostly used to simplify the process of using machine learning to solve general practical problems. However, this approach allows automating the process of comparing the results of various studies and significantly simplifying the process of their generalization. Automated machine learning is based on a predefined set of interdependent processes that are combined into a pipeline. Usually, the pipeline consists of automated operations from data processing till the results generation, that significantly improves the process of results obtaining and analysis. The article describes the main principles and features of the system and how each pipeline's phase was modified to accomplish our task. The described architecture makes it possible to automate such processes as data collection and processing, creation and training of machine learning models, generation and processing of the received outputs. As a result, the user of the designed system can configure any of the supported algorithms using the user interface and later analyze the obtained results using the corresponding program modules.

Keywords

• neural networks
• cryptocurrency rate forecasting methods
• system for methods analysis