Comparison of Performance of Artificial Neural Network (ANN) and Random Forest (RF) in the Classification of Land Cover Zones of Urban Slum Region

Abstract

India is one of the world’s largest economies and economic growth has remained continuous. This has led to accelerating urbanization which requires proper planning and monitoring. As the urban areas are expanding, urban slum areas are also increasing along with it. These growing urban slum areas require proper observation so that existing resources can be employed to provide these regions with the best possible livelihood conditions. For this purpose, urban slum areas as well as surrounding land resources should be well identified and classified so that the existing land resources can be appropriately utilized for future implementation of development activities. Machine learning classification algorithms are found to be very suitable for the identification and classification of remotely sensed images. Their efficiency in feature identification and extraction has established these algorithms as important tools in decision making. In this study, our major objective is to identify and classify different land cover zones in the urban slums areas of Chingrajpara area of Chhattisgarh using remotely-sensed images. For this purpose, high-resolution images, collected using unmanned aerial vehicles (UAVs), are used and these images are classified into different land cover features using two different machine learning algorithms Artificial Neural Network (ANN) and Random Forest (RF). The results obtained show that the overall accuracy achieved by ANN and RF are 72.6% and 84.35% respectively. The study highlights the role and importance of landcover classification for future planning and management.