<![CDATA[Bitcoin price prediction has become an important research topic due to the high volatility and dynamic behavior of cryptocurrency markets. Traditional forecasting approaches often struggle to capture both complex temporal patterns and external market sentiment simultaneously. This study aims to predict the daily closing price of Bitcoin using a Long Short-Term Memory (LSTM) model integrated with Google Trends data as an external sentiment variable. The dataset consists of 1,926 daily records from December 30, 2020, to April 8, 2026, containing Bitcoin historical prices, technical indicators, and Google Trends features. The proposed methodology applies preprocessing, MinMax normalization, technical indicator generation, feature engineering, and Walk-Forward Validation using TimeSeriesSplit with 5 splits to avoid data leakage in time-series forecasting. Three models, namely XGBoost, LSTM, and Hybrid LSTM + XGBoost, were compared to determine the best baseline model. Experimental results show that the LSTM model achieved the best performance, with an average RMSE of 2053.25, outperforming both the XGBoost and the Hybrid LSTM + XGBoost models. However, integrating all Google Trends features decreased prediction performance due to increased noise. Feature selection on relevant Google Trends variables, such as the crypto market and BTC, successfully improved the model’s performance, although it still did not surpass the baseline LSTM model. The study concludes that LSTM is highly effective for Bitcoin price forecasting and that proper feature selection is essential when integrating external sentiment data, such as Google Trends.]]>
Copyrights © 2026
