A Critical Review of Literature on the Application of Machine Learning Algorithms in Cropland Suitability Modelling

Abstract Land Suitability Analysis (LSA) for is a crucial process for optimizing land use, crop yields, and sustainability. Machine Learning (ML) approaches provide robust alternatives to conventional suitability modelling. This study adopted a critical approach to review the literature on the progress, limitations, gaps, and opportunities on the application of ML algorithms in crop farming suitability modelling. Since 2012, Maxent (42%) and Random Forest (RF) (25%) dominate applications. Additionally, MODIS and Landsat are increasingly used for suitability assessment due to spectral and temporal resolution. This study identifies the most sensitive remotely sensed-derived variables (e.g., NDVI, EVI and land surface temperature) that could be used to simulate biophysical environmental factors and highlight the role of geospatial technologies in crop suitability modelling. ML effectiveness is constrained by demand for large, high-quality datasets. Integration with Analytical Hierarchy Process (AHP) and fuzzy logic can address limitations, where AHP can incorporate expert judgment for weighting criteria and fuzzy logic can handle uncertainty and imprecision in environmental data. Socio-economic factors (land tenure, market access) should be integrated to improve reliability. Integration of remote sensing and Internet of Things (IoT) with ML enables real-time crop monitoring.