Evaluation and comparison of eight machine learning models in land use/land cover mapping using Landsat 8 OLI: a case study of the northern region of Iran

Abstract

Land use land cover change mapping has been used for monitoring environmental changes as an essential factor to study on the earth’s surface land cover in the field of climate change phenomena such as floods and droughts. Remote sensing images have been suggested to present inexpensive and fine-scale data offering multi-temporal coverage. This tool is useful in the field of environmental monitoring, land-cover mapping, and urban planning. This study aims to evaluate eight machine learning algorithms for image classification implemented in WEKA and R programming language. Firstly, Landsat 8 OLI/TIRS Level-2 images based on eight machine learning techniques including Random Forest, Decision Table, DTNB, J48, Lazy IBK, Multilayer Perceptron, Non-Nested Generalized Exemplars (NN ge), and Simple Logistic are classified. Then, obtained results are compared in term of Overall Accuracy (OA), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE) for land use land cover mapping. Among the eight machine learning algorithms used for image classification based on the training and test dataset, NN ge classifier is ranked first with values of 100, 0, and 0 for Overall Accuracy, Mean Absolute Error and Root Mean Squared Error respectively. All machine learning algorithms had an Overall Accuracy of more than 99% for the training dataset. On the other hand, for the test dataset, J48 and DTNB algorithms had the worst performance with values of 88.1188 and 76.9802 respectively for the Overall Accuracy.