This study investigates the impact of feature selection on the predictive performance of machine learning models in the Brazilian derivatives market. Precisely, by integrating technical analysis indicators, such as Simple and Exponential Moving Averages and Standard Deviation, derived from intraday price data of mini-Bovespa index futures traded on the Brazilian Stock Exchange, we construct a robust dataset of historical prices from the BovDBV2 database. We employ several feature selection methods to reduce dimensionality, minimize overfitting, and improve model generalization. A Random Forest classifier, evaluated using a 9-fold stratified cross-validation scheme, is employed to compare the performance of models trained on the full feature set versus those trained on reduced subsets. Our findings demonstrate that feature reduction significantly improves accuracy on test data while reducing computational costs. These results provide valuable insights for investors and traders, highlighting the importance of feature selection in developing predictive models capable of adapting to dynamic derivatives market conditions.