Comparing Classification Algorithms for Predicting Spatial Land Cover via Landscape Indices in Nashik, India

The application of robust machine learning (ML) classification algorithms to connect regional land cover (LC) types such as built-up areas, bare lands, vegetation, and water bodies with landscape indices like Normalized Difference Built-up Index (NDBI), Normalized Difference Bareness Index (NDBaI), Normalized Difference Vegetation Index (NDVI), and Normalized Difference Water Index (NDWI) can facilitate the automatic generation of accurate Land Cover Maps. In this study, eight robust ML algorithms were applied, namely Logistic Regression, Support Vector Regression, Decision Trees, Random Forests, Gaussian Naïve Bayes, Gradient Boosting Classifier, K-Nearest Neighbours, and Artificial Neural Network, to establish a predictive relationship between various LC classes and NDBI, NDBaI, NDVI, and NDWI. LANDSAT satellite datasets were used to provide ample training and testing data for the eight classifiers. Hyperparameter tuning and tenfold cross-validation were employed to achieve the highest possible accuracy. Among the eight classifiers analyzed, the Artificial Neural Network stood out, initially achieving 71.23% accuracy in 2.67 min, matching the highest attainable accuracy after 54.78 min of hyperparameter tuning. The Support Vector Classifier started at 69.69%, reaching 70.00% after 2 h and 49 min. Random Forest, starting at 67.85%, notably increased to 70.26% in a 5.58-h tuning period. Logistic Regression, with an initial 68.57%, improved to 68.76% in just 48 s of tuning.