Extended Abstract
Background and Objectives: The hot and humid region of Iran experiences extremely hot summers with high humidity, making it one of the most challenging climates globally. Analyzing the features of vernacular houses in these areas can offer valuable insights for modern housing design. One of the key challenges researchers encounter in architectural typology studies is selecting appropriate case studies. Bandar Kong, a coastal city along the Persian Gulf, features traditional houses with four main components: windcatchers, Sabat (shaded walkways), main rooms (living areas), and yards, along with non-living spaces. Understanding the organization of these elements can help develop a typology of vernacular houses in Bandar Kong.
Methods: One of the key applications of machine learning methods recently employed in architectural research is the measurement of similarity in architectural images. Categorizing and describing architectural features within each category is essential for identifying architectural types. Previous studies have utilized cosine similarity for measuring the similarity of architectural plans. Cosine similarity measurement criterion is particularly effective for evaluating sparse vectors and is commonly used in positive spaces with a range of [0,1]. Due to the diverse nature of architectural data, this method has proven effective for evaluating plan image similarities. The aim of this research is to apply machine learning techniques to select case studies and cluster the houses of Bandar Kong based on the shape and arrangement of windcatchers, sabat, courtyards, and living spaces. For this, Anaconda version 3.9 and Jupiter 6.4.5 were utilized. The cosine distance algorithm was employed to measure similarity in terms of shape and spatial relationships. The hierarchical algorithm, using the average linkage method, was used to extract and categorize the features of each plan.
Findings: According to the analysis, the architectural plans of Bandar Kong houses can be divided in 3 different clusters. Scatter diagrams of each cluster can shows characteristics of each cluster. According to the scatter diagram, the length, width, and height consistently fall within the ranges of 2.5-3.5 meters for length and width, and 9-9.5 meters for height. By analyzing the scatter diagram of the characteristics of each cluster, the following results have been extracted. In the first cluster, the windcatcher is located in the east, and the sabbat or courtyard is located on the west side of it. The main rooms are mostly located on the north side and the service spaces are located on the east and west sides. In the second cluster, the windcatcher is centrally placed on the west side of the house. Here, the plan layout tends to extend along a north-south axis, with living rooms positioned on both the west and east sides. In the third cluster, the windcatcher is located on the west side of the plan. In this category, the extension of the plans is mostly east-west. The northern side of the windcatcher typically features the Gatieh room, and in most plans in this group, the wind room connects to either the northern room or the Gatieh. According to the similarity measurement, the plans of Younesi, Golbat and Karchi houses have the highest shape similarity and spatial relationships with other plans.
Conclusion: Nowadays, with the growing volume of data and the complexity of data analysis, software solutions are increasingly used across various fields, including architecture, to minimize errors. One major challenge in architectural research is the classification and selection of case studies for analyzing architectural types. In this study, after evaluating the shape similarity and spatial relationships of architectural plans, the Younesi, Karchi, and Golbat houses were selected as case studies due to their highest similarity in both shape and spatial relations compared to other plans. Using a hierarchical classification method with average linkage, the plans were grouped into three main categories. The defining characteristics of each category were extracted from the charts and compared with the corresponding case study from each group. As a result, the Karchi house represents the first category, the Golbat house represents the second, and the Younesi house represents the third, with their respective features aligning closely with the extracted characteristics of each category.