This study explores the modelling, use, and analysis of various machine learning and deep learning models for financial data. It focuses on testing the accuracy of cryptocurrency price prediction and analysing and demonstrating various interpretable features of machine learning models. This paper explains the features and limitations of traditional LSTM and CNN models in detail, combines both advantages and uses the CNN-LSTM model as the research model. In order to verify its superiority, the paper chooses the most common BTCUSDT in the market as the research object, and the data of the basic indicators of BTCUSDT from January 2019 to 31 December 2023 are compared by finding the optimal parameters of the model and establishing the CNN model and LSTM model. The experimental results show that the CNN-LSTM model is more advantageous than the traditional model in predicting the price of BTCUSDT and conclude that the proposed CNN-LSTM model has good feasibility and universality in predicting the effect.