Cryptosight: Decrypting Future Trends with LSTM, PyCaret and Jinja 2.0
Soham Motiram Mhatre, Brajesh Kumar Mishra, Ishan Kumar Raja, Bigit Krishna Goswami, Sneha Agrawal, Pranjal Upadhyay · 2024
The primary objective of this research is to develop and evaluate combined cryptocurrency price prediction models using Long Short-Term Memory (LSTM) neural networks along with traditional time series models and machine learning techniques like Random Forest and Linear Regression. The LSTM models will capture long-term dependencies in sequential cryptocurrency price data, while the other models provide complementary predictive capabilities. This multi-model approach aims to leverage the strengths of different techniques to improve overall prediction accuracy. Additionally, this research seeks to explore the practical application of these predictions in real-world cryptocurrency trading scenarios. The study will collect historical price data and relevant news for major cryptocurrencies from reputable sources. Data preprocessing, feature engineering, model training, hyper-parameter tuning, and evaluation using metrics like RMSE, MAE, and R-squared will be performed. The predictive performance of the LSTM-based models will be compared against baseline models, with statistical significance testing. Interpretability analysis will provide insights into the factors driving price movements. Furthermore, the research will investigate integrating the predictions into trading algorithms and evaluate their efficacy through back-testing on historical data. Limitations, challenges, and future research directions in cryptocurrency price prediction will also be discussed.