Supervised Machine Learning Algorithms for Land Cover Classification in Casablanca, Morocco

Abstract

This study embarks on an evaluation of the efficacy of six supervised machine learning algorithms in the classification of land cover in Casablanca, Morocco, utilizing Landsat satellite imagery. Employing the Google Earth Engine (GEE) platform for data collection, the research encompasses meticulous pre-processing steps and the application of various supervised algorithms, followed by a comprehensive evaluation of their performance. The city of Casablanca, characterized by rapid urbanization and evolving land-use patterns, presents an exemplary case for scrutinizing the algorithms' ability to accurately classify different land zones. These zones encompass water bodies, urban areas, agricultural lands, barren terrains, and forests. The algorithms under scrutiny include Support Vector Machine (SVM), Random Forest (RF), Classification and Regression Trees (CART), Minimum Distance (MD), Decision Tree (DT), and Gradient Tree Boosting (GTB). The assessment of classification outcomes leverages multiple accuracy indicators, namely overall accuracy (OA), Kappa coefficient, user accuracy (UA), and producer accuracy (PA). Results indicate that the Random Forest algorithm exhibits superior performance, achieving an accuracy of 95.42%, while the Support Vector Machine algorithm lags with a lower accuracy of 83%. This investigation underscores the critical role of advanced machine learning algorithms in land cover classification, a pivotal aspect for urban and regional planning, natural resource management, and risk assessment in rapidly changing environments.