Cryptocurrency market traders primarily employ fundamental and technical analysis to predict price trends. A crucial component of technical analysis involves the use of technical indicators. However, the effectiveness of these indicators can be hindered by noise and redundancy. This study aims to enhance Bitcoin/USDT price forecasting by applying feature selection to a set of technical indicators. We employed mutual information (MI) to identify the most influential indicators. Subsequently, two ensemble learning algorithms (Extra Trees Regressor, Light Gradient Boosting Machine) were applied to build predictive models. Our findings demonstrate that feature selection using Mutual Information improves model performance. By reducing feature dimensionality and leveraging robust algorithms, we achieved more accurate and reliable price predictions. This research contributes to the advancement of cryptocurrency trading strategies by optimizing feature engineering and model selection processes within the technical analysis framework.