Deep-Based Hybrid Model For Enhancing Land Region Prediction Outcomes

Although satellite data is not a novel approach to mapping wetlands, its accuracy depends on data quality and categorization algorithms, making it a continuous challenge. The increasing accessibility of satellite data and the development of data modelling technologies, such as big data modelling using machine learning (ML) algorithms, have improved the accuracy of this type of mapping. Our work demonstrated a simple, scalable, and reliable wetland classification using linear Support Vector Machine (l-SVM) and unsupervised k-Nearest Neighbor (k-NN) machine learning approaches. Normalized difference vegetation index (NDVI) and normalized difference water index (NDWI), the two primary inputs were modelled using Landsat optical data. Subsequently, l-SVM was used to supervise k-NN data using training data collected from outside observations. The accuracy test results indicate that there is a considerable reduction in error for both optical indices, and all prediction models have an accuracy of about 99%.