AI Based Cryptocurrency Price Prediction: A Comparative Analysis of Traditional & Deep Learning Models with Sentiment Integration

Abstract

Cryptocurrencies have risen as significant intangible assets in the financial landscape utilizing cryptography and blockchain technology to facilitate secure and decentralized transactions. The cryptocurrency market is characterized by high volatility and fast price fluctuations since the introduction of Bitcoin back in 2009. Due to the non-linear and timeseries’ long-term dependencies, traditional financial forecasting models face difficulty in capturing the crazy nature of the crypto market. This thesis highlights this fact in a comparative manner measuring the performance of AI models (LSTM, CNN-LSTM, and Sentiment-LSTM) and traditional model (ARIMA) for predicting cryptocurrency price. Additionally, the impact of sentiment on prediction accuracy is examined. The study uses a quantitative research design, by including historical pricing data and sentiment analysis in model development. The results indicate that AI models perform better in forecasting than traditional models by demonstrating superior performance metrics like MSE, RMSE, MAE and MAPE. Sentiment analysis proved to show a variable impact. For Bitcoin (BTC) and Ethereum (ETH) sentiment analysis showed marginal improvement while for Ripple (XRP), the impact is negligible. The findings contribute to existing literature by providing quantitative evidence on the superiority of AI models in cryptocurrency forecasting and highlighting the potential of sentiment analysis. Limitations of this study included reliance on singles source of social media sentiment and usage of small basket of cryptocurrencies due to budgetary and computational constraints. This research provides investors, analysts, and any stakeholder valuable insights along with the framework to easily navigate the highly volatile cryptocurrency market.