A carregar...
Publicação
Bitcoin Price Prediction: A Comparative analysis of PatchTST and LSTM
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.43 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Transformers-based algorithms have revolutionized the field of natural language processing.
In this context, these types of algorithms are being adapted to time series forecasting. The
present study compares PatchTST, a Transformers-based algorithm, against the commonly
used Long Short-Term Memory (LSTM) algorithm in the problem of Bitcoin forecasting. The
methodology employed in this study involved the collection of data from blockchain,
macroeconomics, and technical indicators. This data was used to predict Bitcoin prices across
various lookback and horizon windows. The models were evaluated using standard evaluation
metrics to ensure robust performance comparison. The results demonstrated that PatchTST
outperformed LSTM in both short and long-term horizons. Specifically, for the short-term
horizon, PatchTST achieved a Root Mean Square Error (RMSE) of 10.36 compared to 149.72
for LSTM. For long-term horizons, PatchTST achieved an RMSE of 1 909.96, whereas LSTM had
an RMSE of 2 310.41. PatchTST exhibits superior capability in capturing and predicting Bitcoin
prices compared to LSTM, making it a more effective algorithm for time series forecasting in
the context of cryptocurrency markets.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Knowledge Management and Business Intelligence
Palavras-chave
Bitcoin Cryptocurrencies Machine Learning Long Short-Term Memory Transformers PatchTST SDG 4 - Quality education SDG 8 - Decent work and economic growth