Models for Analyzing Cryptocurrency Market Dynamics with Behavioral Metrics of Stakeholders Based on Social Media Data

Abstract

The use of data mining techniques in the context of the behavioral economy of virtual assets based on social media data allows for a more accurate assessment of the price movements of cryptocurrencies. The study builds models for forecasting prices in the cryptocurrency market, taking into account the behavioral factors of stakeholders based on social data from the TikTok platform. The data for this study were obtained using social media application programming interfaces. The main stages of the study included data collection, processing and aggregation, scaling and correlation analysis, building and evaluating the relevant models. As a result of the study, key behavioral metrics of social networks are identified. Correlation analysis showed strong linear links between TikTok’s social performance and weak links to the price of bitcoin. The study builds linear models, polynomial regression, Decision Tree and Random Forest. Behavioral metrics such as the number of shares, likes, comments, shares, and views are used. Models are evaluated by testing with use of MSE and MAE metrics. The results suggest the limited effectiveness of linear regression for predicting the prices of cryptocurrencies due to the non-linear nature of the market. The Decision Tree model has shown some success in predicting bitcoin prices, however, deviations in forecasts have increased over time, especially in the face of market fluctuations. Polynomial regression and the Random Forest model demonstrate higher accuracy in forecasts. Based on the comparison of MSE and MAE indicators, the Random Forest turned out to be the most effective model for predicting bitcoin prices among those considered.

Keywords

• cryptocurrencies
• behavior patterns
• api (application programming interface)
• social networks
• machine learning
• behavioral economics
• forecasting models