In this study, we investigated a novel recommendation system that utilizes collective brain activity. Conventional collaborative filtering relies on conscious inputs such as user ratings, which do not necessarily capture subconscious human preferences. In this study, we propose a method that treats brain activity data obtained via near-infrared spectroscopy as a form of collective intelligence, estimates brain activity features for unobserved products through collaborative filtering, and predicts preferences using a support vector machine. Experimental findings confirm the effectiveness of the proposed method, showing only a 9.2% decrease in accuracy compared with the results obtained using actual measured brain activity features. Future enhancements may include integrating deep learning, applying majority voting across multiple models, and adapting the method for binary recommendation tasks. This method offers a promising direction for recommendation systems that incorporate human sensitivity and subconscious responses.