Identification of Urban Slums Using Classification Algorithms—A Geospatial Approach

Abstract

Urbanization is a dynamic phenomenon which plays a vital role in the global scenario and it is estimated to increase in the near future. Rapid urban sprawl attracts rural communities in search of employment and basic amenities. Improved transport network and communication have rendered the rural population to upgrade job prospectus by migrating from outskirts to the city center. This in turn, has raised the density of informal settlements into the urban landscape which is termed as ‘urban slums’. In this study, an attempt has been made to discriminate formal and informal settlements for Chingrajpara, Chhattisgarh by employing various classification algorithms using Unmanned Aerial System (UAS) dataset. Incorporating pixel-based approaches like Maximum Likelihood and Mahalanobis distance classifiers, ensemble decision tree namely Random Forest classifier, back propagation algorithm such as Neural Net classifier and object-based image analysis using feature extraction to geometrically rectified datasets yields classified results with diverse accuracies. Selection of representative training samples favors for acquiring reliable accuracies. This study also addresses the suitable classifier that outperforms for Very High Resolution (VHR) datasets depending on the accuracy assessment. Since UAV data produces excellent resolution images, the land cover feature appears distinct. Among the array of advancements, point clouds provide 3D information that exhibits true ground features. Thus the resultant classified images are validated using elevation information estimated from point cloud datasets. Methodical results serve the urban planners and spatial analysts for systematic designing, thus alleviating random growth of informal settlements as VHR UAV datasets are a boon to the field of geospatial technology.