Optimization of Bitcoin Price and Return Prediction Model Based on CNN-LSTM-KAN

This paper focuson optimizing the prediction of returns of cryptocurrencies, Ethereum and Bitcoins, by machine learning models. Our target is to improve the accuracy of prediction by CNN-LSTM-KAN model, and analyze the market performance of major cryptocurrencies, comparing their investment returns and risks with traditional financial assets such as the S\&P 500 and 10-year Treasury bond yield. All the original historical data are collected from Yahoo Finance, and covering the time period from December 1, 2017, to December 31, 2023. This study mainly use LSTM and CNN-LSTM-KAN models to forecast cryptocurrency returns and compares the results, metrics such as Mean Squared Error (MSE) and R-squared (R) score are used to evaluate the prediction performance of these models. The findings emphasize the relatively optimistic potential for high returns ofcryptocurrencies, aside with great volatility and risk compared to traditional financial assets. The study also explores the effect of these predictions for investors decision making and market regulation, its results suggest that machine learning models can provide some valuable insights into cryptocurrency market dynamics, and help investors in making better decisions.