Random Forest Algorithm for Land Cover Classification

Since the launch of the first land observation satellite Landsat-1 in 1972, many machine learning algorithms have been used to classify pixels in Thematic Mapper (TM) imagery. Classification methods range from parametric supervised classification algorithms such as maximum likelihood, unsupervised algorithms such as ISODAT and k-means clustering to machine learning algorithms such as artificial neural, decision trees, support vector machines, and ensembles classifiers. Various ensemble classification algorithms have been proposed in recent years. Most widely used ensemble classification algorithm is Random Forest. The Random Forest classifier uses bootstrap aggregating for form an ensemble of classification and induction tree like tree classifiers. A few researchers have used Random Forest for land cover analysis. However, the potential of Random Forest has not yet been fully explored by the remote sensing community. In this paper we compare classification accuracy of Random Forest with other commonly used algorithms such as the maximum likelihood, minimum distance, decision tree, neural network, and support vector machine classifiers.