Comparison of machine learning methods for classification of Bandar Kong windcatchers

Hot and humid region of Iran is one of the hardest climates in the world. Due to its proximity to the sea and in order to use of coastal winds, windcatcher is one of the architectural elements of these areas, including Bandar Kong. Classification of architectural types is the first step in understanding the features governing architecture. This research aims to classify the catchers of Bandar Kong using machine learning methods. For this purpose, the plans of Bandar Kong have been categorized in two General ways, based on shape and characteristics of plans and the results have been compared. In the first method, the similarity of 35 windcatchers is calculated using the Cosine Distance method in Anaconda3.9 .each plans is compared 34 times with other plans. In second step plans are are clustered using using Clustmap from Seaborn Library. In the next method, the characteristics of windcatchers such as length, width and location of windcatcher have been extracted from each plan and classified in Anaconda using complete linkage and average linkage methods from Numpy library. Windcatcher plans had been divided to 6, 5 and 4 clusters using different methods. The clusters show that clustering based on images, had placed more similar plans in one cluster.