Use of electroencephalogram and long short-term memory networks to recognize design preferences of users toward architectural design alternatives

Abstract

In this study, we propose an electroencephalogram (EEG)-based long short-term memory networks model for recognizing user preferences toward architectural design images. An EEG is an approach that records the electrical activity in the brain, and EEG-based affection recognition is a technique used for quantitatively recognizing human emotion by analysing the recorded signals. Decision-makers' subjective reactions toward architectural design alternatives may play a key role in the architectural planning and design stage. In this regard, the proposed model enables the quantitative recognition of their preferences and supports architects in the planning and design stages. The suggested model classifies the recorded data using a deep-learning technique. To build the model, an EEG recording experiment was conducted with 18 subjects, who were asked to select their most/least preferred images among eight images of small-housing design. Post recording, a positive and negative affect schedule questionnaire was distributed to the subjects to rate their affection. Google TensorFlow and Keras were used to structure the model. After training, precision, recall, and f1 score metrics were used to evaluate and validate the model. This model can help designers to evaluate design alternatives in terms of decision-making. Moreover, as this model uses biosignal data, which is universal to humans, architectural design processes for children, the elderly, etc., may be supported. Furthermore, a data-driven design database may be proposed in a future research for cross-validating with previous methods such as interviews and observations.