Abstract:
"Cryptocurrency trading heavily relies on the accurate analysis of price trends for profitable decision-making. Manual trend analysis currently proves time-consuming and is likely to be false and then leads to huge losses in assets. The unpredictable nature and high fluctuations of cryptocurrency prices pose a complex forecasting environment, necessitating the development of effective models for accurate predictions. Cryptocurrency markets grow vastly around many users and demand effective tools for price prediction to assist in making profitable decisions.
The project takes a deep learning approach, utilizing Long Short-Term Memory (LSTM) networks, an advanced recurrent neural network (RNN), to forecast prices using historical cryptocurrency charts. The data fed into the LSTM model for training and testing are extracted features from cryptocurrency price charts using a pre-trained convolutional neural network (CNN) model. The dataset is created manually and consists of chart images gathered from online trading view applications and has been used for model training and testing. This approach has been chosen by the author over the current approaches which use numerical datasets to provide traders with a tool they can rely on to view accurately predicted price patterns. Moreover, integrating explainable AI techniques into the model enhances the transparency of the predictions while improving users' trust in the application outcomes.
The model shows a strong performance during training and validation, as evidenced by the remarkably low loss and error metrics. Specifically, the training metrics highlight an impressive degree of accuracy, with a mean absolute error (MAE) of 0.0122 and a mean squared error (MSE) of 2.8953e-04. The validation metrics further affirm the model's reliability, demonstrating even lower error rates, which suggests a high level of precision in predicting outcomes based on unseen data."
