Predicting Architectural Space Preferences Using EEG-Based Emotion Analysis: A CNN-LSTM Approach

Abstract

Architectural elements—such as shapes, colors, and lighting—significantly influence how users emotionally respond to spaces. This study addresses the challenge of capturing unconscious and rapid emotional responses by employing a 32-channel electroencephalography (EEG) approach with 40 participants, who viewed multiple images of architectural spaces while real-time brain activity was recorded. Event-related potential (ERP) analysis focusing on N100, N200, P300, and late positive potential confirmed reliable differences in neural signals between preferred and non-preferred stimuli. Two convolutional neural network long short-term memory deep learning models were trained on the EEG data: one using all the ERP segments, and the other focusing on statistically significant ERP features. The first model achieved a high recall but a relatively lower precision, while the second improved accuracy and precision at the expense of recall. These findings suggest real-time, objective measures of users’ emotional responses can inform early-stage architectural design and reduce reliance on subjective evaluations. By integrating EEG-based insights into smart architecture or virtual reality simulations, designers may optimize building features to align with user preferences and well-being, contributing to the development of effective and user-centric built environments.