COMPARATIVE ASSESSMENT OF MACHINE LEARNING ALGORITHMS FOR LAND USE AND LAND COVER CLASSIFICATION USING MULTISPECTRAL REMOTE SENSING IMAGE

Abstract

In developing countries, rapid and unregulated population growth, along with economic and industrial growth and development have accelerated the rate of changes in land use and land cover (LULC) during the early twenty-first century. One of the most effective ways to assess and manage the land transformation is through quantitative assessment of LULC changes. To find the best classifier for further uses of earth observation data, it is necessary to compare the accuracy of several LULC modelling algorithms. In this paper, four machine learning (ML) algorithms/classifier namely support vector machine (SVM), k-nearest neighbor (KNN), random forest (RF) and decision tree (DT) have been examined using Landsat 8 Operational Land Imager (OLI) sensor data. All the algorithms performed almost in a similar accuracy level, where SVM scores 84.00% of Cohen kappa and 90.74% of overall accuracy. These figures are 82.99% and 90.05% respectively for KNN. For RF, Cohen kappa and overall accuracy recorded 78.24% and 87.62% respectively. DT gained 74.46% for Cohen kappa and 85.51% for overall accuracy. Clearly SVM outperformed other algorithms in low spatial-resolution satellite data. The outcome of the study would help in different stages of LULC modelling. Moreover, it would help the planners and policy makers in formulation and preparation of land-based policy and planning to ensure sustainable living environment.