Evaluating the Architectural Designs Using Machine Learning: The Case of Two Modes

Machine Learning (ML) is one significant subfield of Artificial Intelligence (AI) that impacts most of today’s industries. This research is part of a broader research project that investigates enquiring creativity in machine learning versus the human design processes. The research aims at developing a computational model that evaluates architectural designs using machine learning algorithms, to fulfill project goal of investigating the computer’s ability to generate architectural designs. Two ML algorithms were trained with a set of design drawings along with their own evaluation. Three evaluation criteria are identified; Spaces’ orientation, functional zoning, and usability of area. Each design case is accompanied by a given evaluation; the evaluation is set by practitioner architects per each criterion. To train the ML algorithm, a custom-built drafting tool was developed, the drafting tool accepts a given architectural plan design drawn on an evaluation grid. Each architectural plan drawing is identified in terms of classes that represent space uses. The tool then scans the grid in order to extract the features related to each of the evaluation criteria. Within this research paper the two algorithms were trained with a dataset of 15 architectural designs, and then tested with five test cases that include the five possible grades. Results showed that both ML algorithms have learning curves, and that one algorithm is significantly better than the other in learning the three grading criteria. This research serves a long-term objective to implement ML in generating designs with specified qualities.