Automatic Generation of Architectural Plans with Machine Learning

Abstract

The fact that computer software has no intuition about the design process is the main reason not to outsource that entire process to computers, this study aims to use artificial intelligence solutions that automatically produce architectural plans based on machine learning. The research combines quantitative and qualitative data using genetic algorithms, machine learning (k-means clustering), and instance-based neural networks. The results of this study show that, unlike methods based on a combination of genetic algorithms and genetic programming, it is possible to improve the accuracy and speed of map generation by combining three genetic algorithms, machine learning, and a pattern-based graph neural network. Another feature of the proposed method is a nearly 90 percent learning rate in identifying and presenting complete designs.

Keywords:

• Automatic Plan Generation
• Genetic Algorithm
• Machine Learning
• Graph Neural Network
• Artificial Intelligence

Data Statement

The data supporting this study’s findings are available from the corresponding author, Reza Babakhani, upon request.

Additional information

Notes on contributors

Reza Babakhani

Reza Babakhani, PhD, focuses on research in architecture and artificial intelligence. He is the Director of the Architecture & AI Laboratory for the International Federation of Inventors’ Associations (IFIA).