Application of Machine Learning and Remote Sensing Techniques for Mapping Informal Settlements. Case Study of Cúcuta—Norte de Santander

This study aims to identify and map informal settlements in the city of San José de Cúcuta using Sentinel-2 satellite imagery and geospatial analysis tools available on the Google Earth Engine platform. For this process, a multitemporal composite image was created, enriched with various spectral indices (such as NDVI, SAVI, MNDWI, among others), and a model was trained using supervised classification machine learning algorithms, specifically Random Forest, with representative samples of relevant land cover types in the urban context of the study area. The Validation was performed using 30% of the samples, generating a confusion matrix to evaluate its performance. The results were positive, with an overall accuracy of 87.5% and a Kappa index of 0.86, indicating a high level of agreement between the classified and actual land cover classes. While some classes, such as informal settlements, showed greater confusion, others—such as water bodies and vegetation—were classified more accurately, reinforcing the reliability of the methodology used. Finally, after a spatial post-processing step to remove noise from the composite image, the informal settlement class was extracted to retain only significant groupings of the same class. Based on this final coverage, the total area occupied by informal settlements was calculated, as well as its variability during the 2018–2024 period.