A hybrid approach to urban growth assessment using K-Nearest Neighbor, Support Vector Machine, Random Forest, and Maximum Likelihood (Case study: West Tehran )

Abstract

Urbanization is a growing concern, and satellite images play a crucial role in assessing urban growth. The availability and time series characteristics of satellite images make them a powerful tool for evaluating changes in phenomena. To begin working with satellite images, it is necessary to take samples and classify the images according to the region's complications. Using a specific classification method for time series of images may not produce accurate results to evaluate the changes in a phenomenon, and much depends on the dispersion of the samples taken from the images. In this study, 4 machine learning algorithms (K-Nearest Neighbour, Support Vector Machine, Random Forest (Random Trees), and Maximum Likelihood) were used to classify images from three periods of Landsat satellite imagery (Landsat 7, 8, 9) at two 10-year intervals (2003, 2013, and 2023). In four areas of Tehran (2, 5, 21, 22), this has been applied to urban growth. For the classification results, accuracy and Kappa coefficient were used. Therefore, Using the KNN method with a Kappa coefficient of 91%, Landsat image 7 performed best due to the uniformity of the samples. Additionally, Landsat images 8 and 9 were successfully analysed with the SVM method with an accuracy of 97% and 94%, respectively, as well as a Kappa coefficient of 95% and 89%. Urban growth is also evaluated using selected methods for each image. For this purpose, to evaluate the changes in a specific area, the study area is divided into equal parts, and using zonal statistics, the area of each element of the changes is applied to the divided areas. Therefore, between 2003 and 2013, urban growth was 10%, between 2013 and 2023, it was 24%, and as a result, between 2023 and 2003, it was 34%. Additionally, we examine the change in barren and green lands in this study. Our study offers the most accurate hybrid approach to image classification for urban growth, and it can provide valuable information to urban planners and policymakers for managing urban growth and promoting sustainable development in cities.