Obstructive Sleep Apnea (OSA) is a common disorder that affects quality of life and increases the risk of serious diseases. This study proposes an automatic system for OSA detection based on EEG signals, implementing optimal electrode selection and analyzing the impact of different brain regions on model performance. Using the public ISRUC-SLEEP database, the EEG were preprocessed to extract relevant features and train a supervised learning model. The results show that combining channels from the central and occipital regions provides an optimal balance between accuracy and computational cost (AUC-ROC of 95.72%). Although the configuration using all EEG channels achieved the highest overall accuracy (95.88%), reduced configurations such as F4-O2 deliver good performance (94%) with a 55% reduction in computational cost. This study contributes to the design of accessible and accurate systems for OSA detection, demonstrating the effectiveness of optimal electrode selection while maintaining a balance between accuracy and computational cost.