Comparative Analysis Of The Existing Machine Learning Based Approaches For Land Use And Land Cover Of Geographical Areas

Abstract

Land use and land cover (LULC) analysis is critical for environmental monitoring, urban planning, and resource management. Traditional methods have limitations in handling large datasets and complex spatial relationships, making it difficult to accurately classify and predict LULC changes. Machine learning (ML) has emerged as a powerful tool for LULC classification, offering new methods to improve accuracy and efficiency.   This systematic review, guided by PRISMA principles, compares various ML techniques used for LULC classification. The review focuses on supervised, unsupervised, and deep learning algorithms applied to geographical data. The inclusion criteria ensured the relevance and quality of selected studies, considering only peer-reviewed publications in English that specifically address ML techniques for LULC classification. A thorough search was conducted in academic databases such as Web of Science, Scopus, IEEE Xplore, and others, using specific keywords related to LULC and ML techniques.  A total of 80 articles were selected for an in-depth analysis. The study identifies Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Decision Trees (DT), and Random Forest (RF) as the most prominent ML classifiers for LULC. Each classifier has distinct strengths and weaknesses. SVM performs best in high-dimensional spaces, KNN is effective with low-resolution images, DT provides clear reasoning but risks overfitting, and RF excels in processing diverse data formats.   ML techniques significantly enhance the accuracy and efficiency of LULC classification, with each method offering unique advantages depending on the data characteristics and application requirements. Future research should focus on integrating these techniques with Geographic Information Systems (GIS) and remote sensing data to further improve LULC mapping and prediction