Comparison of Machine Learning Classification Techniques for Change Analysis in Pauri Garhwal District of Uttarakhand, India.

Land use and land cover changes have greatly contributed to global warming and increased urbanization during the last two decades. It is crucial to track alterations of the globe's surface's usage of land as well as cover in order to obtain accurate information about the study region for all development planning scenarios. Land cover categorization is critical for identifying changes, managing development, and tracking and assessing the ground's surface area. Along with RS and GIS methodologies, machine learning algorithms have recently gained popularity in land use and land cover change detection studies. This present study was conducted to detect land use and land cover (LULC) in a heterogeneous landscape of Pauri Garhwal district of Uttarakhand, India by using different machine learning classifiers, i.e. (i) Random forest. (ii) SmileCart (iii) SmileGradientTreeBoost and (iv) SVM. Overall accuracy and kappa coefficient found for RF, SVM, CART and GTB were (0.98, 0.96), (0.97, 0.95), (0.80, 0.75) and (0.89, 0.80) respectively. Results show that Random Forest (RF) provided highest classification accuracy among all four classifier. [ABSTRACT FROM AUTHOR]